Uploaded bydiezotore
14 views

Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys

Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys

Embed presentation

Download to read offline
Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys download https://ebookbell.com/product/visual-object-processing-a- cognitive-neuropsychological-approach-glyn-w-humphreys-52461610 Explore and download more ebooks at ebookbell.com
Here are some recommended products that we believe you will be interested in. You can click the link to download. Visual Object Tracking Using Deep Learning Ashish Kumar https://ebookbell.com/product/visual-object-tracking-using-deep- learning-ashish-kumar-52598712 Visual Object Tracking From Correlation Filter To Deep Learning 1st Ed 2021 Weiwei Xing https://ebookbell.com/product/visual-object-tracking-from-correlation- filter-to-deep-learning-1st-ed-2021-weiwei-xing-36380984 Visual Object Recognition 1st Edition Grauman K Leibe B https://ebookbell.com/product/visual-object-recognition-1st-edition- grauman-k-leibe-b-4584652 The Visual Object Of Desire In Late Medieval England Sarah Stanbury https://ebookbell.com/product/the-visual-object-of-desire-in-late- medieval-england-sarah-stanbury-51968168
Machine Learning Challenges Evaluating Predictive Uncertainty Visual Object Classification And Recognising Tectual Entailment First Pascal Machine Learning Challenges Workshop Mlcw 2005 Southampton Uk April 1113 2005 Revised Selected Papers 1st Edition Joaquin Quionerocandela Auth https://ebookbell.com/product/machine-learning-challenges-evaluating- predictive-uncertainty-visual-object-classification-and-recognising- tectual-entailment-first-pascal-machine-learning-challenges-workshop- mlcw-2005-southampton-uk-april-1113-2005-revised-selected-papers-1st- edition-joaquin-quionerocandela-auth-4239698 Objectoriented Programming With Visual Basicnet Michael Mcmillan https://ebookbell.com/product/objectoriented-programming-with-visual- basicnet-michael-mcmillan-2004060 Beginning Object Oriented Programming Using Visual Basic 2005 From Novie To Professiona Dan Clark https://ebookbell.com/product/beginning-object-oriented-programming- using-visual-basic-2005-from-novie-to-professiona-dan-clark-4112518 Visual Perception And Robotic Manipulation 3d Object Recognition Tracking And Handeye Coordination 1st Edition Geoffrey Taylor https://ebookbell.com/product/visual-perception-and-robotic- manipulation-3d-object-recognition-tracking-and-handeye- coordination-1st-edition-geoffrey-taylor-983282 Visual Saliency From Pixellevel To Objectlevel Analysis 1st Ed Jianming Zhang https://ebookbell.com/product/visual-saliency-from-pixellevel-to- objectlevel-analysis-1st-ed-jianming-zhang-9964488
PSYCHOLOGY LIBRARY EDITIONS: PERCEPTION Volume 15 VISUAL OBJECT PROCESSING
VISUAL OBJECT PROCESSING A Cognitive Neuropsychological Approach Edited by GLYN W. HUMPHREYS AND M. JANE RIDDOCH
First published in 1987 by Lawrence Erlbaum Associates Ltd This edition first published in 2017 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN and by Routledge 711 Third Avenue, New York, NY 10017 Routledge is an imprint o f the Taylor & Francis Group, an informa business © 1987 by Lawrence Erlbaum Associates Ltd All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-1-138-68824-7 (Set) ISBN: 978-1-315-22895-2 (Set) (ebk) ISBN: 978-1-138-20973-2 (Volume 15) (hbk) ISBN: 978-1-138-20976-3 (Volume 15) (pbk) ISBN: 978-1-315-45685-0 (Volume 15) (ebk) Publisher’s Note The publisher has gone to great lengths to ensure the quality of this reprint but points out that some imperfections in the original copies may be apparent. Disclaimer The publisher has made every effort to trace copyright holders and would welcome correspondence from those they have been unable to trace.
Visual Object Processing: A Cognitive Neuropsychological Approach Edited by Glyn W. Humphreys Department of Psychology, Birkbeck College,U.K. and M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, U.K.
Copyright © 1987 by Lawrence Erlbaum Associates Ltd All rights reserved. No part of this book may be reproduced in any form, by photostat, microform, retrieval system, or any other means without the prior written permission of the publisher. Lawrence Erlbaum Associates Ltd., Publishers 27 Palmeira Mansions Church Road Hove East Sussex, BN3 2FA U.K. British Library Cataloguing in Publication Data Visual object processing : a cognitive neuropsychological account. 1. Visual perception I. Humphreys, Glyn W. II. Riddoch, M. Jane 152.1/4 BF241 ISBN 0-86377-045-2 Typeset by Spire Print Services Ltd., Salisbury, Wilts Printed and bound by A. Wheaton & Co. Ltd., Exeter
Contents List of Contributors ix Preface xi 1. Introduction: Cognitive Neuropsychology and Visual Object Processing 1 Glyn W. Humphreys and M. lane Riddoch 1. Introduction 1 2. Cognitive Neuropsychology and Visual Object Processing 2 3. The Book 11 2. Visual Object Perception from a Computational Perspective 17 Kent A. Stevens 1. Introduction 17 2. Modularity in Vision 19 3. Survey of Computational Problems 21 4. Concerning Visual Defects 35 5. Conclusion 39 3. Normal and Pathological Processes in Visual Object Constancy 43 Glyn W. Humphreys and Philip T. Quinlan 1. Introduction 43 2. The Problem of Visual Object Constancy 44 3. Approaches to Object Constancy 48 4. Animal Work 82 5. Neurological Impairments of Human Vision 88 6. Some Conclusions 95 V
4. Picture Naming 107 M. lane Riddoch and Glyn W. Humphreys 1. Introduction 107 2. Studies of Picture Naming in Normal Subjects 108 3. Data from Brain Damaged Subjects 127 4. Conclusions 139 5. Information Processing and Laterality Effects for Object and Face Perception 145 lustine Sergent 1. Introduction 145 2. Information Processing 148 3. Some Properties of the Visual System and their Implications 153 4. Hemispheric Processing Assymetry 162 5. Conclusion 167 6. The Clinical Spectrum and Localisation of Visual Agnosia 175 Andrew Kertesz 1. Introduction 175 2. Varieties of Visual Agnosia 176 3. Agnosia and Cortical Blindness 182 4. Associated Deficits 183 5. The Lesions Producing Visual Agnosia 189 6. The Functions of Visual Recognition 191 7. Apperceptive Agnosia: The Specification and Description of Constructs 197 lohn Campion 1. Introduction 197 2. A Neuropsychological Framework for Vision 198 3. Sensory Loss Theory and Agnosia 200 4. A Case of Apperceptive Agnosia 206 5. Two Other Patients 221 6. Conclusions 227 8. Object Concepts and Object Names: Some Deductions from Acquired Disorders of Word Processing 233 Elaine Funnell '~ 1. Modularity and the Division of Labour 233 2. Conceptual Representations of Objects 236 3. Levels of Object Recognition 239 4. Semantic Processing and Imageable/Concrete Word Concepts 243 5. Shared Conceptual Information for Object Names and Objects 253 9. Dementia and Visual Agnosia 261 Oscar S. M. Marin 1. Dementia: Definition and Nosology 261 2. Perceptual Disorders in Dementia 264 V¡ CONTENTS
CONTENTS 3. Dementias as a Model for Neuropsychological Research 4. The Nature of Perceptual Disorders in Dementia 268 5. The Spectrum of Visual Agnosic Syndromes 275 10. The Fractionation of Visual Agnosia 281 Glyn W. Humphreys and M. Jane Riddoch 1. Introduction 281 2. Apperceptive Agnosia 282 3. Associative Agnosia 292 4. Classification and Lesion Sites 294 5. Re-classifying Visual Agnosia 299 6. Conclusions 304 Author Index 307 265 Subject Index 315 vii
List of Contributors John Campion Behavioural Science Division, Admiralty Research Establishment, Queens Road, Teddington, Middlesex, U.K. Elaine Funnell Cognitive Neuropsychology Research Group, Department of Psychology, University of London, Malet Street, London WC1E 7HX, U.K. Glyn W. Humphreys Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street, London WC1E7HX, U.K. Andrew Kertesz Clinical Neurology Department; St. Joseph's Hospital, 268 Grosvenor Street, London, Ontario N6A 4V2, Canada Oscar S. M. Marin Laboratory of Cognitive Neuropsychology, Department of Neurology, Good Samaritan Hospital and Medical Center, Portland, Oregon 97210, U.S.A. Philip T. Quinlan Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street; London WC1E 7HX, U.K. M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, Romford Road, Stratford, London E15 4LZ, U.K. Justine Sergent Department of Neurology, McGill University, Montreal Neurological Institute, 3801 University Street, Montreal PQ, H3A 2B4, Canada Kent A. Stevens Department of Computer and Information Science, University of Oregon, Eugene, Oregon 97403, U.S.A. V.O.P.—A* ix
Preface This book has its origins in a meeting of the European International Neuropsychology Society (INS) at Aachen in June 1984. We had been asked to organise a symposium on visual agnosia at the conference. Our idea was to use the symposium to try to bring together workers from various disciplines in order to throw new light on the old problem of how visual object processing occurs, and on how it can be disturbed by damage to the brain. We invited a number of contributors to the book to give talks. The meeting seemed to go well, and there was interest expressed in the idea of using converging approaches to understanding object processing. Some time later, the thought occurred to us that a book covering these converging approaches might also serve some purpose, and this is the final product. In the period since the INS conference, work on all aspects of visual object processing has advanced considerably, and undoubtedly it will continue to do following the publication of this book. Our hope is to provide a state-of-the-art summary of work on the topic, and also to convey something of the excitement of working in a field subject to rapid development. The final editing of the book has been helped by many people: all the “ support staff” at Lawrence Erlbaum, the careful (though anonymous reviewers), and the Cognitive Neuropsychology Research Group at Birkbeck College, especially Philip Quinlan and Cathy Price. To all these people, our thanks. We are also grateful to Alan Cowey, Peter Dodwell, Geoff Hinton, Stephen Lupker, Stephen Palmer, and W. H. Freeman for xi
x ii PREFACE permission to reproduce figures. All the photographs were taken by Freddie Elliot, and Daphne Ring and Sue Godfrey provided much-needed last-minute typing. Our final thanks go to Iain, Alec, and Kate, for putting up with our working in the evenings!
To Glyn and Dorothy
1 Introduction: Cognitive Neuropsychology and Visual Object Processing Glyn W. Humphreys Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street, London WC1E 7HX, U.K. M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, Romford Road, Stratford, London E15 4LZ, U.K. The development of an efficient visual system has clearly been important for our survival. The ability to represent the position of objects with respect to our body position allows us to negotiate the world efficiently, so that we can avoid obstacles and reach appropriately for those objects we wish to grasp. The ability to recognise that an object has the same structure when seen from different viewpoints, with different lighting, etc. enables us to assign it the same response under a variety of contrasting viewing conditions. The ability to associate the viewed object with our memory facilitates such diverse activities as choosing food, taking early evasive action from predators, and selecting group members from amongst a crowd. Humans are also able to assign agreed names to objects, and so can communicate information about objects in an efficient manner. All of these different abilities reflect aspects of visual object processing. Normally, visual object processing is so effortless and (seemingly) immediate that it is difficult to conceptualise the complexity of the proces- ses involved. However, some inkling of this complexity can be gained when one encounters patients who, following brain damage, have selective dif- ficulties in such apparently simple abilities as matching objects presented from different viewpoints, knowing how to use a common object or assigning a name to it. Such cases suggest that visual object processing involves a set of rather independent abilities, each of which can be selectively affected by brain damage. 1. INTRODUCTION 1
2 HUMPHREYS & RIDDOCH In this book, we have attempted to bring together work by researchers concerned with the functional and neurological mechanisms underlying visual object processing, and the ways in which such mechanisms can be neurologically impaired. We have termed it a “ Cognitive Neuropsycholog- ical” approach, because we believe it tries to relate evidence from neurological impairments of visual object processing to models of normal performance in a new and important way. Two broad aims are apparent. One is to test models of normal performance by evaluating how well the models account for the patterns of impairment and preservation of abilities that can occur following brain damage. The other is to use models of normal performance to further our understanding of acquired disorders of visual object processing. These aims distinguish the approach from neuro- psychological work whose primary aim is to relate acquired deficits to the sites of damage, and from work in the field of cognitive psychology which attempts only to develop models of normal performance (see also Coltheart, 1984, 1985). 2. COGNITIVE NEUROPSYCHOLOGY AND VISUAL OBJECT PROCESSING The characteristics of acognitive neuropsychological approach to behaviour are only currently developing and, to some extent, they will be shaped by the nature of the particular field of study. In at least some areas, progress has been limited by the lack of well-specified information processing models (see Coltheart, 1985). It seems clear that we will remain unable to explain or to predict the various ways in which a given ability may break down if we have little idea of the component processes involved. Recent developments in the field of visual object processing, however, provide grounds for optimism in this respect. In particular, the work on computer vision by David Marr and his associates (Marr, 1976, 1980, 1982; Marr & Nishihara, 1978) has produced a more developed informa- tion processing account of visual object processing than was hitherto available. Marr suggested that visual object processing could be considered in terms of a series of processing stages, each of which produces a more elaborated representation of a visually presented object, making different types of information explicit (see, for instance, Chapter 3 by Humphreys & Quinlan). Each of these processing stages operates as a separate module, which makes information available for subsequent stages but which is to a large extent functionally independent of those stages. Marr also outlined certain procedures which may determine how information at each stage is represented and transformed to provide the input for subse- quent processing stages. Perhaps by considering the types of representation mediating visual object processing, and the procedures which create or
1. INTRODUCTION 3 enable access to these representations to occur, we may gain some insights into the disorders of object processing which are produced by brain damage. Also, instances of selective problems following brain damage may provide constraints on the models of normal performance. These, then, are the general aims of a cognitive neuropsychology approach to visual object processing. However, the application of this approach to vision has not proved so straightforward. There are probably several reasons for this. 2.1. Applying Theories from Computer Vision to Humans To begin with, there is an initial issue concerning the optimal way to “frame” the problems involved in visual object processing. As noted earlier, some of the most explicit models of the component processes involved are derived from the field of computer vision. Typically, such models are not aimed at accounting for how processing is carried out by humans. It follows that data from either normal observers or from subjects with acquired disorders of vision cannot be used to evaluate such models in their own right; the data may only be used to evaluate the models as accounts of human vision. The question arises as to which models are then appropriate. For instance, an axiom of computational approaches to information processing is that theories may be expressed at different levels of abstraction (see Stevens, Chapter 2). At a very high level of abstraction (usually termed a computational level), theories may be concerned only with what the requirements of the system may be. Thus Marr (1979) states that “ vision (requires) the construction of efficient symbolic descriptions from images of the world” . Descriptions may be constructed which make available information which is unique to the object seen from a given viewpoint (i.e., some form of viewpoint dependent description), enabling us to manipu- late the object in space. Descriptions may also be constructed which enable the object to address stored information irrespective of the viewpoint (i.e., some form of abstract object description), so that recognition can be achieved across different views (see Sutherland, 1979). According to this approach, vision requires the construction of each of the above kinds of description. Another level of theory would be concerned with how these descriptions are derived (the level of the algorithm), and yet another with how any processes are implemented in hardware terms (the level of the hardware; see Marr, 1982; Stevens, Chapter 2). We need to be clear about which level of theory is being addressed by any piece of behavioural evidence. Evidence that visual object processing in humans is highly constrained by viewpoint (see Humphreys & Quinlan, Chapter 3) would
be pertinent to computational-level theories of the kind outlined by Marr (1982); evidence concerning the kinds of procedures used in constructing different types of visual description, however, would be pertinent to theories at the level of algorithm, not to computational level theories. The rejection of a theory as applied to human vision at the level of algorithm (e.g., in terms of the type of procedure used to generate a particular visual description) may not entail the rejection of the theory at a computational level (in the sense that that description may still mediate performance). Thus, the kinds of theory we accept will be dependent on the level of description to which any evidence applies. 2.2. Independence between Levels of Description The argument that there exist separate levels of description of any complex information processing problem makes sense in computer science, where the constraints on one description of a task (e.g., in terms of a flow-chart or a description of the algorithm) may be quite independent of those at another (such as hardware). It is not clear that this is true of the brain. Details concerning the speed of neuronal transmission, the specialisation of different visual areas, the prior tuning of cells within any visual area, etc., may be crucial in determining the manner in which different processes are carried out. If this suggestion is correct, then we may not be able to apply a theory at one level of description without considering its implementation at other levels in some detail. In cognitive neuropsychological approaches to other areas, particularly to the study of language and reading, evidence concerning the nature of the hardware has been effectively ignored (e.g. Coltheart, 1985; Mehler, Morton & Jusczyk, 1984). One reason for this selective view is that our knowledge of the brain structures mediating tasks such as reading has been insufficient to constrain theories concerned with reading mechanisms. This state of relative ignorance contrasts with that in the field of vision, where we now have quite detailed accounts of the neural systems involved, of their preferences for different types of visual input, and of their mapping (e.g. Cowey, 1985). It is possible that our developing knowledge of the neural hardware of vision will help determine theories of visual processing mechanisms. To illustrate this point, consider evidence concerning the selective preferences of cells in different visual areas of the cortex for different properties of visual stimuli. Recent work suggests that at least some cells in area V4 of the monkey cortex (see Fig. 11, Humphreys & Quinlan, Chapter 3) exhibit preferences for specific colours even when the wavelength of the illumination on a surface changes (i.e., they exhibit colour constancy; Zeki, 1980, 1983). Such cells might well be involved in colour identification (cf. Desimone, Schein, Moran, & Ungerleider, 1985; 4 HUMPHREYS & RIDDOCH
Land, 1977). Theories concerned with how we identify, say, a letter in a particular colour may need to take account of the possibility that the neural structures subserving colour identification may differ from those finally involved in letter identification. For instance, such theories may need to posit that colour and form identification are functionally independent (cf. Ballard, Hinton, & Sejnowski, 1983; Treisman & Gelade, 1980). It would follow that in order to identify a particular letter in a particular colour (e.g., against a background consisting of the same letter in a different colour and different letters in the target colour), some special mechanism may be required to co-ordinate information from the different visual areas, such as linking by common retinotopic locations (cf. Quinlan & Humphreys, 1987) or by the serial application of focal attention (Treisman & Souther, 1985). The process of interaction between the different levels of description need not be one way, however. Higher-level accounts of the mechanisms involved in visual object processing may also constrain theories at the hardware level. For instance, there is some evidence to suggest that the detection of boundaries between different visual textures can be deter- mined by activity in sets of detectors responding to simple local, properties of stimuli, such as colour and line orientation (cf. Beck, 1967, 1982; Treisman & Gelade, 1980). These properties correspond to discrimina- tions which may be expected from the activity of colour or orientation- tuned cells in the visual cortex (cf. Hubei & Weisel, 1968). Other behavioural evidence indicates that texture boundaries may also be distin- guished on the basis of more complex visual information, such as “ elon- gated blobs” and the number of line terminators (Julesz, 1980, 1981), or (possibly) closure (Treisman & Paterson, 1984). Given the argument that texture boundaries are signalled by specific hardware units tuned to the discriminating properties, it should follow that there exist detectors for the above complex types of information (see Julesz, 1980, 1986). Inspection of those theories currently shaping cognitive neuro- psychological work on vision indicates a fairly wide range of approaches to issues concerning the relations between different levels of theoretical description (although these differences are rarely made explicit). For example, Marr’s (1982) account of the processes leading up to the genera- tion of a three-dimensional (3-D) representation of a shape assumes that these processes may be described independently of their neural implemen- tation. Other processing accounts seem to be more intimately linked with the known properties of neural hardware. Most “ connectionist” theories of visual object processing at least purport to take seriously implications for the design of a system where complex knowledge is expressed via activity in a set of simple (neurone-like) computational units (see Hinton & Ander- son, 1981; Humphreys & Quinlan, this volume; McClelland & Rumelhart, 1. INTRODUCTION 5
1986). Some connectionist theories not only allocate separate processing units for specific visual properties of stimuli (such as lines at particular orientations and positions; see Fig. 8, Chapter 3, Humphreys & Quinlan, this volume), they also group units corresponding to properties within a given dimension (such as colour or size) to form a “ parameter space” , where the value(s) of a stimulus along that dimension is specified (cf. Ballard et al., 1983; Feldman & Ballard, 1982). Such parameter spaces may be considered analogous to areas of the visual cortex selectively tuned to different visual dimensions. What are the implicationsof these different approaches for the aims of cognitive neuropsychology we outlined earlier? First, consider how we may use evidence drawn from brain damaged patients to assess the different models. Models which maintain a clear independence between descriptions of processing mechanisms and hardware descriptions can only be addres- sed by information concerning the functional deficits of patients (i.e., by the type of behavioural deficit exhibited and by its relation to other component abilities). This position is akin to that typical of cognitive neuropsychological approaches in other areas, such as reading mentioned earlier. Models in which processing mechanisms are tied to the properties and locations of specific cortical areas may similarly be addressed by functional evidence. However, in addition, such models are also open to evaluation by evidence concerning the site and extent of damage. For instance, consider the possibility that a patient continues to exhibit colour constancy following selective ablation of area V4. This result would be difficult to accommodate for a theory which posited a colour-processing system explicitly modelled on our knowledge of the properties of cells in V4, and which maintained that colour-constancy could only be achieved using that colour-processing system. A theory positing a colour-processing system which does not have a one-to-one relationship with area V4 would not, of course, be similarly damaged. Accounts of brain damaged patients in terms of these different approaches will also differ. Models assuming independent algorithm-level and hardware-level descriptions account for behavioural deficits in patients in terms of patterns of dysfunction in (abstracted) processing components. Models assuming some explicit relations between the levels of description may additionally take on board information about the nature of the brain damage. Again, this may be illustrated by reference to theories positing that a given ability reflects patterns of activation across representations tuned to specific properties of stimuli (the connectionist accounts discussed by Humphreys & Quinlan, Chapter 3; the cascade model discussed by Riddoch & Humphreys, Chapter 4). Such theories may (eventually) offer different accounts of a particular deficit according to whether the deficit is due to a single selective lesion, due to a pattern of disseminated lesions (cf. 6 HUMPHREYS & RIDDOCH
1. INTRODUCTION 7 Campion, Chapter 7) or due to a reduced density of connections within a neuronal network (cf. Marin, Chapter 9). Taken to their furthest conclu- sion, such theories may even return to the approach of classifying patients according to the nature (i.e. the type, extent, and site) of their brain damage, at least where our knowledge of the nature of the brain damage and its relation to behaviour is sufficiently precise. Campion, Chapter 7, argues for this approach to classifying patients with visual processing deficits following carbon monoxide poisoning. Whichever approach is adopted, we believe it important to realise that models can differ both in terms of their processing mechanisms and in terms of their assumptions concerning how the hypothesised processing mechanisms relate to the hardware of the system. The kinds of inferences we make will depend on the assumed processing— hardware relations. 2.3. Processing Modules The assumption that information processing is modular, in the sense that it is thought to comprise of a set of functionally independent processes, is one that is clearly of fundamental importance to cognitive neuropsychological approaches to behaviour (see Coltheart,1985). Such an assumption is at the heart of accounts of brain damaged patients in terms of a pattern of intact and impaired component abilities (see Funnell, Chapter 8). A principle of modularity has also been espoused by workers in computer vision (Marr, 1976; Stevens, Chapter 2). In the latter context, workers have suggested that breaking a large computation up into a series of separate modules is useful because a small change to one process will not have consequences on others. By specifying a complex behaviour in terms of a series of separate processing modules, direct links between computer vision theories and theories of human behaviour seem to be facilitated. However, it is relevant to point out that concepts of modularity can differ. One view, outlined by Funnell (Chapter 8), maintains that the principle of modularity relates to the knowledge requirements and proces- ses demanded by different tasks, not to the tasks themselves. When this approach is applied to brain damaged subjects, we would expect that damage could selectively affect certain types of knowledge or certain processes. Two tasks will dissociate only to the extent that they tap the same damaged or intact processing modules. It is also possible to outline a rather different interpretation of modular- ity, apparent amongst at least some computer vision researchers. For instance, Stevens (Chapter 2) considers the view that any two computa- tions may be thought of as being modular to the extent that they can be performed independently of one another. One implication of this view is that two tasks may be considered modular even though both tasks require
*he same type of process. Stevens cites the example of surface interpola- tion. Our ability to see a smooth surface at a different depth to its background when presented with a random-dot stereogram (Julesz, 1971) requires the interpolation of this surface between corresponding points in the left and right retinal images. Also, our ability to derive 3-D form information from motion requires the interpolation of a smooth surface between points and edges in the image where velocity profiles can be computed. It is possible that two-surface interpolation procedures are instantiated in two different modules: One concerned with deriving 3-D form information by stereopsis and one concerned with deriving 3-D form information from motion. The job of such modules would be to make available (independently), 3-D form descriptions from different types of stimulus information. Such descriptions would then be combined via a further, subsequent process. The implications of this second view of modularity for understanding neurological deficits in performance are quite different to the implications of the first. Here, we would expect brain damage to selectively affect abilities such as the computation of 3-D form from stereo information, or even a process such as surface interpolation in stereopsis. However, extensive damage to the “ stereopsis module” may nevertheless leave the “form from motion module” (including its surface interpolation component) intact. Present knowledge of the modules mediating visual object processing is probably insufficient to decide between these two views. Indeed, it may even be that the type of modularity governing early visual processing differs from that governing “ higher-order” processes such as recognition and identification. The implications of the differing views of modularity should, however, be borne in mind when considering the kinds of approaches taken when investigating selective visual processing disorders. 2.4. Interactions between Modules In discussing the principle of modularity, Marr (1982) notes that this principle “ does not forbid weak interactions between different modules” . The problem here from a behavioural point of view is in defining the extent and nature of any interactions. At one extreme we might presume that processing is organised in a series of discrete stages, with information being made available to each subsequent stage only following the resolution of processing at an earlier stage. Such discrete stages can be described as being strongly modular, in the sense that each processing stage is immune from other stages (e.g., from the influence of partial information from earlier stages or from information fed forward from later stages). At another extreme, one could imagine a system in which no processing stage is immune from the influence of any other. This can be said of “ interactive 8 HUMPHREYS & RIDDOCH
1. INTRODUCTION 9 activation” models of information processing, in which partial information is transmitted between processing stages and in which later stages feed forward to constrain earlier stages (McClelland & Rumelhart, 1986). Such models are only weakly modular. For instance, brain damage could selec- tively isolate an earlier process from a later process, even though in normality the processes would interact (see Patterson & Morton, 1985, for a similar distinction concerning models of visual word pronunciation). Again, models differ in terms of whether they take a relatively strong or weak view on this topic. Marr’s (1982) view is characteristically strong; the connectionist models discussed by Humphreys and Quinlan and the cas- cade models of picture naming discussed by Riddoch and Humphreys (both in this volume) are relatively weak. Different accounts of brain damaged patients may follow from the contrasting assumptions about modularity. For instance, a strongly modular account of picture naming might hold that the structural characteristics of objects are processed fully before later information about, say, an object’s name is contacted. Accord- ing to this account, a patient who makes visual errors in naming pictures would be thought to have a problem in processes up to the stage of accessing structural knowledge about objects. An account assuming only weak modularity (e.g., the cascade model; Riddoch & Humphreys, Chap- ter 4) would not necessarily take the same position, since stages subsequent to that involving access to structural knowledge about objects could be affected by partially activated structural information. A patient could make visual errors due to problems in these later stages (see Riddoch & Humphreys, 1987, for a more detailed discussion). Additionally, the inferences we make about normal processes, using evidence from patients, are influenced by assumptions about modularity. In a strongly modular system, selective damage to one process would leave any functionally independent processes intact. In a weakly modular system this may not occur, since a problem in one process could constrain the operation of others. Such a conclusion is possible whether one’s account is pitched at an algorithm-level description or at a description at the level of the hardware. Sergent (Chapter 5) is concerned with the functions of the left and right cerebral hemispheres in object perception (a hardware-level issue). She argues that the capabilities of the left and the right hemisphere to process high and low spatial frequency information may differ. For example, the right hemisphere may preferentially operate on lower spatial frequencies than the left. Following unilateral right hemisphere damage, a patient may fail on a task not because the processes required for perfor- mance are functionally lateralised in the right hemisphere (e.g., the proces- ses required to match objects across viewpoints; cf. Warrington, 1982; see also Humphreys & Quinlan, Chapter 3), but because right hemisphere damage changes the quality of the information available to the left hemis-
10 HUMPHREYS & RIDDOCH phere. In normality, task performance may be critically dependent on the co-activation of information in both hemispheres. In this case, we would be wise to be cautious when using single pieces of evidence from brain damaged patients to infer the localisation of one behavioural function either within a single processing module or neural structure. For instance, our conclusions about the nature of any processing module or its anatomi- cal locus would be bolstered by converging evidence that a deficit was functionally localised: by using tasks tapping other processing stages, by examining the temporal characteristics of processing in more detail, or by simulation. 2.5. Positive Attributes The problems involved in taking a cognitive neuropsychological approach to visual object processing thus concern issues about the kinds of descrip- tion that are most appropriate for explaining behaviour, about the relations between the different descriptions, about the kinds of processes which serve as the functional modules within the processing system, and about the relations between modules. Overall, it seems likely that visual object processing will best be understood by detailed examination of different kinds of description and of the contrasting questions which arise. As our knowledge about performance expressed at these different levels increases, we may hope for increasing convergence between such apparently contras- ting fields of study as computer vision, experimental psychology, and neurophysiology. The possibility of constraints between these different fields represents one of the most exciting areas of future development, and one of the strengths of the cognitive neuropsychological approach. One of the probable consequences of developments in the field is that we will produce more detailed models of the component processes involved in different tasks. Such models in turn produce more detailed predictions of the kinds of performance dysfunction which may follow brain damage. Disorders specific to visual object processing are typically classed within the syndrome of visual agnosia, of which two forms are distinguished: apperceptive agnosia, thought to pertain to the perceptual representation of the visual world, and associative agnosia, thought to pertain to the association of perceptual information with world knowledge (see Kertesz, Chapter 6). However, as our knowledge of visual object processing from other areas grows, it seems increasingly unlikely that a simple two-stage account of processing and its breakdown will suffice. The initial “ nosolog- ical” distinction between apperceptive and association stages (cf. Marin, Chapter 9) will fractionate, to provide a more precise account of the similarities and differences between patients. By bringing together work from different research fields, we hope to indicate the ways in which new
1. INTRODUCTION 11 distinctions about visual object processing in general are developing (e.g. Stevens, Humphreys & Quinlan, Sergent, Riddoch & Humphreys, Funnell, this volume), and the ways in which such distinctions are being applied to understanding agnosia (e.g., Kertesz, Marin, Humphreys & Riddoch, this volume, Chapter 10). The utility of the multi-stage view which appears on the horizon (see Kertesz, Humphreys & Riddoch, this volume) can only be judged in time, in terms of how well it accounts for patients and of how useful its distinctions are for accounting for normal performance. In the meantime, it may serve as a fresh focus for work in the area. 3. THE BOOK The book is divided into two sections. In the first section, the chapters approach visual object processing from a consideration of which processes may need to be adopted either to develop an intelligent artificial visual system or to explain normal performance. Stevens (Chapter 2) discusses object processing from the perspective of computer vision. He outlines the kinds of assumptions adopted by workers in this field concerning functional modularity and the implementa- tion of the knowledge into procedural rules within the modules (such as “ rigidity” assumptions in deriving 3-D form information from motion). He goes on to consider some of the implications of computer vision for understanding both normal vision and its breakdown following brain damage. The following chapters by Humphreys and Quinlan (Chapter 3) and by Riddoch and Humphreys (Chapter 4) focus on particular operations in visual object processing: Object constancy and the processes involved in accessing stored structural, semantic and name information about objects. Humphreys and Quinlan’s chapter draws on work from a number of fields, such as computer vision, experimental cognitive psychology, animal vision, and neurologically impaired human vision. After considering the kinds of logical problems which constrain theories of object constancy, evidence concerning the mechanisms involved is reviewed. Humphreys and Quinlan propose that the problems of object constancy are not solved in any single way by the visual system, but that different solutions are adopted depend- ing on the kinds of information available and the uses to which that information must be put. They suggest that future research would prosper by de-emphasising general purpose solutions to constancy and by consider- ing in more detail stimulus and task constraints. Riddoch and Humphreys (Chapter 4) are concerned with the higher- order processes involved in accessing stored knowledge for picture naming. Various accounts of picture naming in normal observers are outlined, and a framework is developed which holds that picture naming operates in
cascade. The cascade model is then applied to account for different forms of picture naming disorder in patients. This chapter provides an example of a “weakly modular” account of visual object processing (see Section 2.4 above), and it attempts to explain both why such an account might be needed and its implications for understanding neurologically impaired patients. The final chapter in this section of the book, by Sergent (Chapter 5), focuses primarily on the anatomical localisation of processing mechanisms in the left and right cerebral hemispheres. However, in attempting to do this, Sergent draws on evidence from normal observers and animals in addition to brain damaged humans. It again provides an illuminating example of how we may look to evidence from seemingly different areas to further our understanding of vision. Necessarily, the chapter highlights the issue of-face perception, since considerably more detailed work on hemis- pheric differences in face perception has been conducted than on such differences in object perception (particularly in terms of developing the kinds of process models necessary to explain both abilities). Nevertheless, many of the lessons to be learned are the same, since faces are perhaps but one extreme example of objects which require accurate and integrated visual information for recognition to occur (see Diamond & Carey, 1986). The importance of viewing normal performance as the product of interac- tions between different processes is emphasised (see Section 2.4). The second section of the book deals with the types of visual object processing disorder which may occur in human subjects after brain dam- age. Kertesz (Chapter 6) outlines the historical background of work in this area. He describes which behavioural deficits are typically labelled as apperceptive or associative agnosia, or as other forms of visual object processing disorder such as optic aphasia; he considers the relations between these disorders and some associated deficits (such as colour and face perception, word recognition, etc.); and he considers the utility of a two-stage account of agnosia (see earlier). Kertesz’s review provides a useful introduction to agnosia from a clinical perspective. Campion (Chapter 7) focuses on the visual processing disorder typically labelled as apperceptive agnosia. His particular concern is with how psychological constructs (such as “perceptual representation” ) should be operationally defined to help understand the deficits in such patients, and on whether distinctions between “ sensory” and “ perceptual” processes are useful in this respect. He discusses how, in at least some instances, it might be useful to group patients according to the type of neurological damage sustained, rather than on the basis of higher-order constructs. Insights might then be gained by considering how the neurological damage con- strains information processing. This approach stresses how “ hardware- level” descriptions can further our understanding of the processing mechanisms mediating task performance. 12 HUMPHREYS & RIDDOCH
1. INTRODUCTION 13 Funnell (Chapter 8) discusses more central deficits in visual object processing, particularly those concerned with the involvement of semantic knowledge. She outlines a novel approach which uses evidence from word matching, naming, and comprehension tasks with patients to address the issue of the nature of the semantic information mediating both these tasks and visual object processing. The relations between our stored knowledge of different properties of objects (such as sensory and functional know- ledge) are also considered, using data drawn from the ratings of normal subjects. Views of the nature of the semantic system are clearly important for understanding central deficits in visual object processing. Funnell argues that the semantic representations of objects differ in terms of the types and number of attributes which are specified, and that the loss of particular types of attribute may be important in understanding certain types of dysfunction. Central deficits in visual object processing are also of concern to Marin (Chapter 9). He deals with the physiological changes to the brain which occur in various forms of dementia, and with how such changes may precipitate deficits in object recognition and naming. Of particular interest is the fact that dementia is often consequent on quite different physiologi- cal changes to those observed in patients with brain lesions following head injury, stroke or tumour. The question is raised whether accounts of the processing deficits sustained also need to reflect these physiological differ- ences (see Section 2.2). Irrespective of whether one is sensitive to such hardware-level effects, cases of dementia can provide examples of remark- ably spared islets of ability amongst otherwise gross behavioural distur- bances; and it seems likely that work with such cases will be increasingly important for understanding the breakdown of what are, in normality, highly-interactive processes. In the final chapter we (Humphreys & Riddoch) attempt to examine in detail some of the implications for classifying visual agnosia in terms of a multi-stage theory of visual object processing. An expanded classification scheme is outlined to account for documented cases of agnosia, and to predict further ways in which object processing might break down. The aim is to provide a more flexible classification scheme which enjoys a better fit with theories of normal vision and which can be used predictively. Consid- eration is also given to the relations between different behavioural deficits and different types and sites of lesion. We hope that the differing fields drawn upon in this volume will attract readers from contrasting backgrounds and that each will find something of interest, not only from within their own field. We also hope that the approach adopted illustrates some of the ways in which we may, in future, cross the boundaries of different disciplines to gain a fuller understanding of human behaviour.
14 HUMPHREYS & RIDDOCH REFERENCES Ballard, D. H., Hinton, G. E., & Seinowski, T. J. (1983). Parallel visual computation. Nature, 306, 21-26. Beck, J. (1967). Perceptual grouping produced by line figures. Perceptual & Psychophysics, 1, 371-394. Beck, J. (1982). Texture segregation. In J. Beck (Ed.), Organization and representation in perception. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Coltheart, M. (1984). Editorial in Cognitive Neuropsychology, 1, 1-8. Coltheart, M. (1985). Cognitive neuropsychology and the study of reading. In Μ. I. Posner and O. S. M. Marin (Eds.), Attention and performance, Vol. XI. Hillsdale, N.J. Lawrence Erlbaum Associates Inc. Cowey, A. (1985). Aspects of cortical organization related to selective attention and selective impairments of visual perception. In M.I. Posner and O.S.M. Marin (Eds.), Attention and performance, Vol. XI. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Desimone, R., Schein, S. J., Moran, J., & Ungerleider, L. G. (1985). Contour, color and shape analysis beyond the striate cortex. Vision Research, 25, 441-452. Diamond, R. & Carey, S. (1986). Why faces are and are not special: An effect of expertise. Journal of Experimental Psychology: General, 115, 107-117. Feldman, J. A. & Ballard, D. H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254. Hinton, G. E. & Anderson, J. A. (1981). Parallel models o f associative memory. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Hubei, D. H. & Wiesel, T. N. (1968). Receptive fields and functional architecture of monkey striate cortex. Journal of Physiology, 195, 215-243. Julesz, B. (1971). Foundations of cylopean perception. Chicago: University of Chicago Press. Julesz, B. (1980). Spatial nonlinearities in the instantaneous perception of textures with identical power spectra. Philosophical Transactions of the Royal Society, London, B290, 83-94. Julesz, B. (1981). Textons, the elements of texture perception and their interaction. Nature, 290, 91-97. Julesz, B. (1986). Texton gradients: The texton theory revisited. Biological Cybernetics, 54, 245-251. Land, E. H. (1977). The retinex theory of color vision. Scientific American, 237, 108-128. Marr, D. (1976). Early processing of visual information. Philosophical Transactions o f the Royal Society, London, B275, 483-524. Marr,D. (1979). Representing and computing visual information. In P. H. Winston and R.H. Brown (Eds.), Artificial intelligence: An MIT perspective. London: MIT Press. Marr, D. (1980). Visual information processing: The structure and creation of visual representations. Philosophical Transactions o f the Royal Society, London, B290, 199-218. Marr, D. (1982). Vision. San Francisco: W. H. Freeman. Marr, D. & Nishihara, H. K. (1978). Representation and recognition of the spatial organiza- tion of three-dimensional shapes. Proceedings o f the Royal Society, London, B200, 269-294. McLelland, J. L. & Rumelhart, D. E. (1986). Parallel distributed processing: Explorations in the microstructure ofcognition, Vol. 2: Psychological and biological models. London: MIT Press. Mehler, J., Morton, J., & Jusczyk, P. W. (1984). On reducing language to biology. Cognitive Neuropsychology, 1, 83-116.
Patterson, Κ. Ε. & Morton, J. (1985). From orthography to phonology: An attempt at an old interpretation. In Κ. E. Patterson, J. C. Marshall, and M. Coltheart (Eds.), Surface dyslexia: Neuropsychological and cognitive studies of phonological reading. London: Lawrence Erlbaum Associates Ltd. Quinlan, P. T. & Humphreys, G. W. (1987). Visual search for targets defined by combina- tions of color, shape, and size: An examination of the task constraints on feature and conjunction searches. Perception & Pyschophysics, 41, 455-472. Riddoch, M. J. & Humphreys, G. W. (1987). Visual object processing in optic aphasia: A case of semantic access agnosia. Cognitive Neuropsychology, 4, 131-186. Sutherland, N. S. (1979). The representationof three-dimensional objects. Nature, 278, 395-398. · Treisman, A. & Gelade, G. A. (1980). A feature integration theory of attention. Cognitive Psychology, 12, 97-136. Treisman, A. & Paterson, R. (1984). Emergent features, attention and object perception. Journal of Experimental Psychology: Human Perception and Performance, 10, 12-31. Treisman, A. & Souther, J. (1985) Search asymmetry: A diagnostic for preattentive proces- sing of separable features. Journal o f Experimental Psychology: General, 114, 285-310. Warrington, E. K. (1982). Neuropsychological studies of object recognition. Philosophical Transactions o f the Royal Society, London, 298, 15-33. Zeki, S. M. (1980). The representation of colours in the cerebral cortex. Nature, 284, 412-418. Zeki, S. M. (1983). Color coding in the cerebral cortex: The responses of wavelength- selective and color-coded cells in monkey visual cortex to changes in wavelength composi- tion. Neuroscience, 9, 767-781. 1. INTRODUCTION 15
Visual Object Perception from a Computational Perspective Kent A. Stevens Department of Computer and Information Science, University of Oregon, Eugene, Oregon 97403, U.S.A. 1. INTRODUCTION David Marr introduced a computational approach towards the study of biological information processing, which stresses that a complex system such as the visual system is feasibly understood only across several levels of abstraction (Marr, 1982; Marr & Poggio, 1977). The suggestion is that no single level of description, be it behavioural, psychophysical, computa- tional, or neurophysiological, can adequately describe both function and implementation with sufficient detail and scope to allow the necessary breadth of understanding. The necessity for multiple levels of description is accepted in computer science without question. When designing or analysing a complex informa- tion processing system one attends either to questions of what the require- ments, goals, and tasks of the system are, or to questions of how the system achieves these goals, such as what the algorithms are and how the data are organised.1 One shifts between these two levels of description often *An algorithm, as defined in computer science, is the formal specification of a particular computation. Informally, it is a sequence of step-by-step instructions or commands which, if interpreted by some processor, would result in the achievement of some computation. An algorithm is expressed by means of a language having an agreed-upon set of constructs for referring to data structures, operations on data structures, and constructs for controlling the execution of these operations. In principle, one can define parallel as well as sequential algorithms, for a variety of types of processing mechanism ranging from analogue to digital 17 2
without regard for a third, more detailed level, that of the actual hardware on which the algorithms run and the data reside. These three levels of description are rather distinct in computer science, and each has its appropriate use and limitations. It is simply not feasible to describe a complex computation in terms of events at the level of the hardware. The technical language and formal principles that are relevant at the level of the digital hardware are neither appropriate nor relevant at the level of algorithm and program, and vice versa. Descriptions of computations are best made via a technical language and a set of formal principles concerned with algorithms, data structures, and other relevant abstractions. It would be theoretically possible to describe the precise pattern of hardware events arising within a computer during, say, the computation of a least squares fit algorithm. The determinism of the digital computer allows, with considerable tedium, the complete specification of its detailed behaviour down to the most elemental state transitions within its electronic components. However, consider working upwards, inferring what the processor is doing, what function it is performing, purely on the basis of the observed hardware event. It is not merely a matter of the overwhelming complexity which makes this not feasible, it is the inappropriateness of the technical language for other than the description of local events. Different languages have been developed for describing computations at different levels of specificity. Of course, computer science is a synthetic science involved with the development and study of computational machinery as it can be humanly conceived and understood. One might argue that limitations in our ability to conceive of and understand complex systems fairly forces one to attend exclusively to either the higher (algorithmic) level or the lower (implementation) level. Be that as it may, one might also argue that computer science has found a fundamental principle of computation (a law perhaps as fundamental as entropy or the conservation laws in physics), namely, that one can describe a computation independently of how it is implemented, or equivalently, that there are arbitrarily many equivalent implementations of a given computation. This 18 STEVENS (discrete state). When one forwards a particular algorithm one suggests a particular recipe for how the computation proceeds and in what form the data to be manipulated are represented. There are infinitely many equivalent algorithms for achieving any given computation (con- sider how many ways one could rewrite a computer program which would not effect its external behaviour). The choice of algorithm depends, in part, on issues of how the given primitive representations and functions are implemented—the machinery at hand. It also depends on design goals such as fault tolerance or “graceful degradation” of behaviour as input is impoverished, time and space efficiency, the immediacy with which the output of the computation is provided to subsequent computations, and so forth (Marr, 1982, p. 106). Thus, when discussing a computation at the level of algorithm one uses rather specific notions of data structure and the operations that transform input into output.
principle, established by Turing and others earlier in this century, has provided the underpinnings for the formal description of computations. Marr suggests that there is no fundamental reason why this principle does not apply equally to biological information processing, and to vision in particular. That is, vision is achieved by processes that extract and manipulate visual information. These computations, while certainly not implemented in the same manner as those performed in silicon technology, are computations nonetheless. Furthermore, being computations, they are amenable to a formal description. One question, then, is what advantage is gained by describing human vision formally as computations? Another question is whether that formal description is really able to be divorced from the biology, or whether it is a nicety of computer science applied to computers, but not of computer science applied to organisms. I think that the answers to both are to be found in the computational work that has been performed in the last decade or so. The purpose of this chapter is to indicate the utility of this approach and to demonstrate where the extensions of these results will take us. 2. MODULARITY IN VISION If vision is to be characterised in terms of information processing problems, how do we determine what any specific problem is in order to study it? Consider colour perception, stereopsis (the creation of depth information from the disparities between the images in the left and right eyes) and the extraction of three-dimensional (3-D) information from motion, all of which have clear computational goals. In principle, one could imagine an independent solution to each, and behaviourally, there appears to be substantial independence of function. The dogmatic view of functional modularity would be that if two computations are theoretically orthogonal (i.e. logically able to be per- formed independently), they should be treated as distinct even though they may share many common underlying mechanisms. For example, the per- ception of colour seems orthogonal to the problem of determining the direction of motion in the retinal image. Consider then the colour- contingent motion and after-effect. One defers the question of how or why such couplings occur until later, when questions of visual mechanism are examined. A reasonable sceptic might regard this approach as dangerously simplis- tic. How can we trust our intuitions as to what computational problems are distinct and which are inherently interrelated, and how can we tell which visual phenomena reflect the visual algorithms and which are merely due to “ implementation details” ? More generally, how can we tell whether our 2. VISUAL OBJECT PERCEPTION 19
decomposition of vision into modules reflects that which nature chose? The still more sceptical might further suggest that it is misguided to imagine any modular decomposition of vision, that vision can be understood only by embracing its complexity and interactions as a whole. I would remind them that we have already developed computer systems the complexity of which defy understanding except in terms of levels of abstraction (ignoring the hardware when thinking of the functionality, and vice versa) and only very locally (considering only a very narrow aspect of the system under the simplifying assumption that there are no important dependencies on global states of the system that are not already reflected by local parameters). Of course, we return to the fact that computer systems are artefacts that are intentionally made with a high degree of strict modularity—we cannot make complicated systems otherwise. Marr (1982, p. 102), however, suggests that nature has also adopted the principle o f modular design, primarily because if the components of a system are not made as independent of one another as the individual tasks allow, a small change to one component will have undesired consequences elsewhere. A system that is not modujar would be difficult to improve or extend; evolution presumably favours a functionally modular design. Functional modularity does not mean strict segregation of function at the neural level, however. Actual segregation or isolation of processes, in fact, has its disadvantages, primarily poor economy of design and the problem of communicating functionally useful information among processes. If two problems are interrelated (such that the solution to one partly constrains the solution to the other), it would be disadvantageous to isolate them. Moreover, just as evolution favours functional modularity, it probably is limited in how literally the modularity can be extended. That is, it would seem improbable that evolution would extend the capabilities of a visual system by implementing the new perceptual ability with neural structures that are strictly isolated from the neural mechanisms that preceded it. More likely would be either the duplication or subdivision of existing cortical structures in order to perform some new, ecologically advantage- ous, computation on existing internal information structures (see, for example, Livingstone and Hubei, 1984, where they suggest that new functional architecture to support colour vision was introduced into the centre of the h]/percolumns found in striate cortex). As a consequence, the new functional architecture would not only be intimately tied with the earlier architecture, it would probably incorporate facilitatory interactions between related processes in an ad hoc and case-by-case manner. I would doubt that evolution adheres to strict design standards for adding new functional capabilities; it is probably governed primarily by the particular problems at hand. For example, consider the problem of combining the 3-D information 20 STEVENS
2. VISUAL OBJECT PERCEPTION 21 delivered simultaneously by stereopsis and motion. Both motion and stereopsis have limitations in the quality of the 3-D information they can deliver individually. In fact, they are complementary so that the results of each process can be combined to provide information that is otherwise ambiguous (Richards, 1985; Waxman & Ullman, 1985). It is not known, however, whether the combining of motion and stereopsis is achieved at a later stage of depth processing, with no “ cross-talk” between the two processes that deliver depth information, or concurrently by more intimate interaction between the two. Since there are, in addition to motion and stereopsis, several additional sources of 3-D information, it would seem easier to imagine a computational architecture for combining 3-D informa- tion that minimises interactions between or among the contributing “ mod- ules” . As to what global architecture is incorporated in human vision is still largely a matter of speculation. Most of the insight has been gained at the level of the brain’s ‘‘microarchitecture’’, as it were, the processing achieved by individual neural units selective to, for example, motion or stereo disparity. 3. SURVEY OF COMPUTATIONAL PROBLEMS In the following we discuss several problems central to spatial perception: stereopsis, the 3-D interpretation of motion, the interpolation or “ filling in” of smooth surfaces, the description of surface shape, the 3-D interpre- tation of contours and texture gradients, and very briefly sketch the processing that might follow the perception of visible surfaces. In examin- ing a variety of information processing problems such as these one can discern a recurring theme in this type of work— one of discovering the physical (or geometrical) constraints that allow the solution to a given visual problem, postulating the method by which the constraints are incorporated into some computation, making explicit various aspects of how the data are represented and processed, and eventually accounting not only for psychophysical results but tying these suggestions to particular neuorophysiological mechanisms. It is a relatively recent idea to attempt to bridge between computationally motivated theories and clinical findings. After the following survey we will close on that note. 3.1. The Matching Problem in Stereopsis A particularly clear problem is presented by the matching problem in stereopsis, namely the determination of the correspondence between left and right images in order that the angular disparity between corresponding elements may be registered. Each image point in the one eye might correspond to any of a range of image points in the other eye, depending on where the point projects when viewed from the perspective of the other
eye. This has also been termed the false-targets problem. For instance, given an image with 4 points in the left eye and 4 points in the right eye, there are 16 possible matchings, only 4 of which are correct, the remaining 12 being false targets. How is the correct matching established over a rich, detailed image where false targets abound? Various proposals could be made as to the image features that are brought into correspondence— intensity points, segments of edges or lines, and more complicated features involving colour, contrast magnitude, and so forth. For example, one might propose local correlation of the intensity values to estimate the stereo disparity in each vicinity. However, is the computational problem of stereopsis being addressed appropriately in such proposals? Does intensity-value correlation solve the underlying computa- tional problem? Empirically, correlation of individual pixels (intensity samples) has generally proven inadequate. The problem lies in the approp- riateness of the tokens that one seeks to match across left and right images. Simply put, there is no reason to expect a pixel-to-pixel correspondence across images. In contrast, Marr and Poggio (1977) recast the computa- tional problem of stereo matching more as one of matching corresponding physical surface features, or at least, of matching elements in the image that have a more reliable tie to surface features than mere pixels. One consequence of this view was to rethink the basis for the matching computation and to derive three constraints on the matching problem. One isthecontinuity constraint. Because the visual world isdominated by smooth, op ^ue surfaces, stereo disparity varies smoothly almost everywhere, be ause the distance to the visible surfaces varies continuously except at surface boundaries. Note that it is precisely at surface boundaries where the correlation approach would fail. Additionally, Marr and Poggio recog- nised two additional constraints on the matching problem—compatibility, which means that the descriptive elements in each eye must be compatible with being adjacent views of the same physical features (hence the contrast signs must match, and edges must match edges, etc.), and uniqueness, which means that the correspondence must be one-to-one almost every- where (the exceptions occurring along occlusion boundaries). Marr (1982, p. 114) reflects on these constraints and suggests the following fundamental assumption o f stereopsis. If a correspondence is established between physically meaningful primitives extracted from left and right images of a scene that contains a sufficient amount of detail, and if the correspondence satisfys the three matching constraints, then that correspondence is physically correct (and it follows that the correspon- dence is unique). Marr suggests that these three constraints are sufficient to solve the stereo matching problem. Identifying, justifying, and demonstrat- ing the sufficiency of such a set constraints is the central activity of deriving a computational theory. We will see other examples momentarily. Note 22 STEVENS
2. VISUAL OBJECT PERCEPTION 23 that while the constraints seem reasonable and plausible it is not clear that much headway has been made on the problem at hand, namely explaining how the visual system solves the stereo matching problem. After all, one must get back to worrying about matching corresponding items in the two images, and terms such as “ physically meaningful” do not immediately suggest particular matching primitives. In fact, it is when one attempts to become more specific that one inevitably goes “ down” a level of abstrac- tion, down to the level of algorithm. Marr and Poggio have examined two quite different algorithms for stereopsis (Marr, 1982), and Mayhew and Frisby (1981) independently proposed similar, highly plausible algorithms vis-à-vis human stereopsis. The insight that produced these computa- tional theories of human stereopsis came from thinking about the underly- ing physical constraints that might, in principle, admit a solution. It should be pointed out that a variety of computational theories for matching have been proposed, each based on a different combination of constraints. Mayhew (1982) describes five constraints that are more or less independent and which can lead to different matching schemes having different performances. The constraints are: (1) the opacity constraint; (2) the epipolar constraint, that a plane passing through the optical centres of the eyes projects to a straight line in each eye, leading to strategies for searching along particular loci for matches; (3) figurai continuity, that edges of surfaces and surface markings are spatially continuous; (4) an object in a given point in space will project to luminance changes in the two eyes with similar descriptions; and (5) the ordering constraint, wherein the stereo projection almost always preserves the ordering of luminance changes along corresponding paths. Mayhew (1982) then shows how different combinations of these constraints have been incorporated in different theories and discusses the relative power of these constraints (in enforcing particular matches or excluding certain matches locally). A computer implementation can be made of a particular algorithm which embodies a given set of these matching constraints, and its behaviour examined when given digitised images of stereo pairs. It is possible to digitise the image with sufficiently fine resolution and broad dynamic range to match the granularity and range of the retinal receptors, then filter the digital image by local lateral-inhibition operators that model retinal gang- lion X-cells. The receptive field of these operators is modelled as a difference of Gaussian distributions (see Marr, 1982; Marr & Hildreth, 1980; Wilson & Giese, 1977). With an implemented model of the early intensity processing and for the specific image features of tokens that are matched between left and right images, one can implement and test a given stereo matching algorithm. The performance of the algorithm can be compared with human vision across a wide range of visual stimuli, from random dot stereograms to natural images. The “ psychophysics” of a
24 STEVENS computer implementation of a stereo matching theory is rather difficult to interpret. If the implementation utterly fails to achieve the same stereo correspondence as do humans in viewing a particular stereo pair one can surmise that the two methods differ in some combination of the matching constraints or the matching tokens, (for example, Mayhew and Frisby, 1981, devised an ingenious stereo pair that tested a detailed prediction regarding the stereo matching tokens incorporated in human vision). It is more difficult to interpret the behaviour of an implemented theory when it is reasonably close to human behaviour but deviates consistently and significantly. One interpretation of the performance disparity is that the human visual system embodies a close, but not exact, engineering approx- imation to the process laid out in the given theory. That is, the theory is correct but does not capture all the implementation details. The alternative explanation is that one does not yet have the right theory, of course. There have been several stereopsis implementations by different investigators, each based on similar but slightly different combinations of computational constraints. The evolution of theories has led to increasingly close matches of predictions of the known psychophysics of stereopsis, but a conclusive demonstration of the validity of a given theory solely on the basis of similarity of behaviour has remained illusive. The basic problem of stereopsis, recall, is determining the correct match between the two images, so that stereo disparity may be computed. There are subsequent computational problems associated with the interpretation of the disparity information, foremost being the computation of distance from disparity (see Mayhew & Longuet-Higgins, 1982). Another problem, which has been considered primarily in the context of stereopsis, but which likely has wider scope, concerns the interpolation or reconstruction of a smooth surface between the points where stereopsis is able to provide depth information. In a sparse random dot stereogram, for instance, once the stereogram is fused, the points appear in space to lie on an apparently continuous surface that exists between the dots, as if the dots were embedded in atransparent sheet. Because interpolation phenomena can be seen also in surfaces depicted by motion sequences, we will return to it as a computational problem in its own right, after considering problems involved in the perception of space from motion. This interweaving of computational problems is commonplace and probably inherent to the nature of vision, and while it makes it difficult to develop clearly demar- cated theories, it is possibly unavoidable. Let us turn to the next topic of our survey, the 3-D interpretation of motion across the retinal image. 3.2. Motion Interpretation Problems The stereo matching problem was one of the first computational problems in vision to be addressed in detail. Another computational problem which has recently received at least as much attention concerns the 3-D interpre-
2. VISUAL OBJECT PERCEPTION 25 tation of motion. Gibson (1950) wrote of the spatial information provided by motion parallax, and showed that in the instantaneous velocity field cast on the retina there is information about the layout of surfaces relative to the observer. That information has generally been regarded as either distance or local surface orientation. What earlier was regarded as merely motion parallax has been decomposed more recently into many computa- tional problems, depending on whether one is concerned with continuous vs. discrete motion, orthographic2vs. perspective projection, and so forth. (For the case of continuous motion in perspective projection see Koen- derink & van Doorn, 1976 and Longuet-Higgins & Prazdny, 1980; for continuous motion in orthographic projection see Hoffman, 1982; and for case of discrete motion in orthographic and perspective projection see Ullman, 1979.) Koenderink and van Doom’s (1976) work demonstrated how the optical velocity field3 contains information about the spatial distribution of sur- faces about an observer as one moves. The local velocity field in the image was shown to be decomposable into quantities that are related to depth, surface orientation, and velocity. Ullman (1979) examined the minimal information from which 3-D structures could be derived, and, for instance, was able to show that three distinct orthographic views of four non- coplanar points in a rigid configuration are sufficient for the recovery of their spatial arrangement, up to a reflection. That is, rigidity of motion results in a unique interpretation under those viewing circumstances. Again, assuming rigidity, Tsai and Huang (1984) have shown that, with certain exceptions, two perspective views of seven points are also suffi- cient. Various other results have been shown for restricted sorts of motion, such as motion about a planar or fixed axis (Bobick, 1983; Hoffman & Flinchbaugh, 1982), which have relevance to the interpretation of articu- lated systems such as animals in locomotion. Rigidity provides the strongest mathematical constraint on the spatial interpretation of motion. In fact, much of the computational work in the last decade has concentrated on the fundamentals of interpreting motion in the abstract, as essentially a mathematical issue. It assumes that motion information is available in some usable form in an image, such as the instantaneous velocity field (retinal angular velocity provided for all visual directions). Whereas Gibson (1950) discussed the wealth of spaital informa- tion provided by the velocity field, the information in the idealised optic Orthographic, or parallel projection is approximated when the variation in distance across a region of a 3-D scene is small relative to the overall distance. 3 The 2-D array of velocity vectors across the retinal image induced by movement relative to the visual surrounds. The velocity field is presumably computed by motion detectors that deliver information about speed and direction of movement across the retinal image. The velocity field is then expected to be a rich source of information for subsequent 3-D analysis.
array that is incident on the retina is quite difficult to extract or “ register” . It appears that computing the velocity field poses substantial problems in its own right, problems that certainly were not foreseen prior to recent computational experiments. For instance, the so-called “ aperture prob- lem” states that only the velocity component normal to an edge or line segment is locally computable, the component tangent to the edge or line cannot be determined by a motion detection operator having a limited aperture or receptive field (Marr & Ullman, 1981). In actuality perhaps only a rather sparse and perhaps crude approximation to the instantaneous velocity field is neurally determined, which would then pose severe prob- lems on any motion processes that use this velocity information as input. The perceptual processes that seek to make 3-D interpretations must be able to cope with errorfu1 and incomplete velocity information (see, for example, Ullman & Hildreth, 1983). While progress has been made regarding the measurement and interpre- tation of motion, the result of the 3-D interpretation, whether represented in terms of depth, surface orientation, or some other form, is still quite obscure. Part of the problem is that we have, as yet, little insight into how spatial information, however extracted, is organised within the human or animal visual system. Some of the issues that must be clarified are: precisely how distance information is encoded, whether distance is made explicit for all visible points across the seen surfaces up to some resolution, how surfaces are described over different scales (for instance, some natural surfaces are smooth on a coarse scale but quite irregular on fine scale, others are smooth in detail but jagged overall). These issues arise when attempting to actually account for the interpretation of motion in natural images, since the irregularities of the seen surfaces result in irregularities in the velocity field, just as smoothness of the surfaces results in smoothness in the velocity field. But the problems of surface description are not restricted to motion interpretation, and that leads us to the next point. 3.3. Surface Interpolation Problems It was mentioned earlier that stereopsis does not provide stereo disparity information at all points in the binocular image, but only at discrete places such as edges, lines, and points that have been brought into stereo correspondence. The intervening, featureless, regions can only be assumed to lie at similar depths if the surface is assumed to be smooth. To make a particular reconstruction of a smooth surface consistent with the points where the disparity is known is not a problem unique to stereopsis, however. Just as stereopsis brings only certain features into stereo corres- pondence, motion provides spatial information only at places for which velocity information can be attributed, such as edge and line segments, 26 STEVENS
2. VISUAL OBJECT PERCEPTION 27 blobs, and points. Nonetheless, one can derive from a discrete pattern of moving dots the impression of seeing a coherent sheet of dots in 3-D, as if one attributes motion to the intervening regions between such features. To do so requires an assumption of viewing a smooth surface when the dots behave in a coherent manner. The same holds for surfaces suggested by contours, such as shown in Fig. 2.1. The smooth surface is depicted by discrete, separated contours, between which no surface information is explicitly available. The impression of an overall surface under the con- tours may be attributed to a surface interpolation process. Some computational attention has been paid to the problem of surface interpolation, the process of reconstructing a smooth surface between the discrete places where the surface is known (see, e.g., Grimson, 1981). Two fundamental research concerns are: (1) whether or not the human visual system explicitly interpolates an entire smooth surface in parallel across the image, and if interpolation exists in human vision: (2) to determine the interpolation strategy or strategies. Although it is not certain that explicit interpolation does take place in human vision, it has been assumed so, and work thus far has centred largely on formalising mathematical schemes for surface reconstruction analogous to the engineering practice of determin- ing smoothly faired surfaces. For instance, one can regard the interpolation as a smooth membrane that is stretched over the discrete places where the surface shape is known, then compute that particular surface shape for the membrane that minimised its elastic potential energy. Rather than use a membrane, an alternative approach is to model the surface as a thin plate, which is resistant to flexure and torsion, and which smoothly bends without fractures or creases to conform with the boundary constraints. Clearly one of the issues that must be dealt with theoretically is how a surface can be described simultaneously as being smooth overall but jagged or irregular in detail, or vice versa. The answer to this might result in different approaches to surface interpolation. One central question, it seems, is to what extent the visual system goes to the effort to explicitly compute multiple surface interpolations at various scales, the alternative being the computation of a surface discription, on demand as it were, for a particular scale of description, and only locally to a given patch of surface. A complementary problem to the interpolation problem is the discon- tinuity detection problem. That is, given a collection of discrete spatial points lying on different, spatially separated surfaces, how does one decide where to fit smooth surfaces and where to postulate surface boundaries? What constitutes a discontinuity, such as a crease or step edge between two smooth surfaces clearly depends on the scale one has in mind (e.g., a crumpled, then flattened sheet of aluminum foil presents both smoothness and jaggedness depending on the scale of reference). But consider the presumably simpler problem of deciding, for a given scale, whether or not
28 STEVENS the surface is creased or smooth at a given point. One approach, for surfaces modelled as thin plates, is to locally measure the “ strain” on the surface and explicitly create discontinuities where the strain becomes excessive. 3.4. Describing Visible Surfaces and Objects Interpolation and the determination of surface creases and boundaries, as well as several other issues such as the description of local surface topogra- phy, are issues that span a broad range of surface perception processes. Parsimony would suggest that the solution to these problems are provided by central mechanisms of surface discription that are common to stereop- sis, motion, shading, texture, and so forth. Marr (1982) collectively termed the representation and associated processes that manipulate internal descriptions of visible surfaces the 2J-D sketch. The name suggesting that it is a sketch or representation of the salient spatial properties of the seen surfaces, not a full 3-D representation of the objects including the enclosed volumes and portions that are not visible from the particular viewpoint. The 2í-D sketch would achieve the integration of 2-D information from different sources such as stereopsis and motion, resulting in an at least locally consistent description of the surfaces that are visible at a given instant. It would also subserve the description of seen objects for purposes of object recognition, manipulation, and so forth. The quantitative description of object and surface layout had been central to computer vision efforts (see, Horn, 1977). Questions of 3-D representation arise in a very concrete manner when one seeks to describe viewed objects within a computer so that they might be automatically recognised, or grasped by a manipulator, or navigated around. However, unlike the proposed 2J-D sketch, in computer vision the representation question has concerned description of a single type of 3-D information derived from a single source (typically either shading or stereopsis) and questions of consistency with information generated from other processes, such as motion, have not been addressed. In contrast, from a psychological perspective there has been concern for the integration of visual cues, notably Attneave’s (1972, 1982) discussions of viewer-centred4 local descriptions of surface orientation and depth, and the intimate relationship between the representation of 3-D information and its extraction from 4A viewer-centred description is relative to the instantaneous perspective the observer has of the surrounding surfaces. Local surface shape and disposition is described relative to the line of sight to each surface patch, so that if the observer moves relative to the surface all values within the viewer-centred description might be expected to change. While volatile, such a description would be the basis for eventually deriving more stable object-centred descriptions (see discussion of axis-based schemes, p. 30).
2. VISUAL OBJECT PERCEPTION 29 images. However, the 2|-D sketch proposal, and perhaps more impor- tantly its theoretical motivation, constitutes the broadest statement of the purpose of early vision, namely the processing of image information in order to develop a description of the visible surfaces (Marr, 1982). Vision achieves descriptions of the visual world, and towards that goal the 2|-D sketch would be one of the important representations within which these descriptions would lie. When phrased in such terms, one also sees clear parallels to Gibson’s (1950) theory that the visual system registers spatial information by means of “ higher order” variables in the optic array (the interested reader is referred to Ullman, 1980, for an excellent discussion of Gibson’s direct perception viewed from this perspective). The major contribution of proposing that early vision culminates in a description of the visible surfaces is probably that it causes one to question precisely what consti- tutes that description. There are many ways to quantify spatial information that are formally equivalent, but presumably only a few such ways are actually used within the visual system to encode such information. At some point in the course of investigation of surface and object perception these issues will have to be pinned down. Making precise and specific the nature and structure of the internal representations of surfaces and objects is extraordinarily difficult. The problem is difficult for many reasons, some of which are sketched in the following. First, one faces the thorny issue that there are indeed many equivalent ways of achieving an equivalent description, given that one cannot rule out specialised processes that access and manipulate the information contained in the representation. To illustrate, suppose the visual system has a 2-D buffer containing perceived depth as a function of visual direction (i.e. a depth map), and associated with it, the ability to compute the gradient of depth in a given locality. Since the magnitude and direction of the gradient of depth is straightforwardly related to the local surface orientation (see Stevens, 1983b), by having the ability to compute the gradient at any point one has, in effect, a local surface orientation map for all smooth surface points. Since a depth map facilitates the perception of local surface orientation, how then does one determine whether or not local surface orientation is made explicit within some 2-D buffer or simply computed locally “ on demand” ? The visual system is likely to be quite robust in its ability to derive different types of spatial inference as required by the visual task at hand. It is still an open question as to what primitive information is stored in a visual buffer and what information is subse- quently inferred or computed on that basis. Depth and local surface orientation constitute but two means for repres- enting surface shape. Other types of 3-D shape primitive can be proposed, such as measures of the local surface curvature (predicates indicating
qualitative flatness or perhaps measures of the sign and magnitude of the Gaussian curvature and the principal directions). More extended topo- graphic features such as ridges and troughs, peaks and dips, etc., might be involved in the description of surface shape in human vision. A ridge or trough or an extremum in surface curvature is clearly visually salient, and it is reasonable to suspect that such features of surface shape are explicitly represented. From a computational point of view it is important to ask how a surface shape should be described, keeping in mind that the representa- tion of surface shape must subserve certain processing tasks that follow, such as object recognition, manipulation, and navigation. In the case of object recognition one of the important criteria of a representation (Marr & Nishihara, 1978) is how stable the description is over changes in viewpoint. Clearly the visual system should derive the same conclusion as to an object’s identity over as many different view- points as possible, with recognition difficulties arising for only particular vantage points where critical features are obscured, such as major axes of elongation or symmetry that are severely foreshortened (see Marr & Nishihara, 1978). Marr and Nishihara (1978) describe a hierarchical representation of objects that is particularly suited for articulated objects such as animals that can be regarded as the composition of parts. In this scheme each part of an object is described in terms of how the cross-section of the part varies as a function of position along its axis (an approach introduced by Binford, 1971). The connectivity among the component parts of an object is then described in terms of how the axes join, the relative sizes of the parts, etc. Marr and Nishihara (1978) observe that such descriptions capture the essence of a broad range of animal shapes, and demonstrate this fact with effective renditions of different animals by pipecleaners. That approach is most effective for describing objects made of connected parts that can themselves be described by axes. More recently, Hoffman (1983) has considered how to decompose a complex surface shape into component parts. The problem can be solved in part by considering the silhouette outline, and identifying the points along the silhouette where the parts intersect (see also Marr, 1977, concerning the interpretation of silhouettes, and the sketch of that work in Section 3.5.). Hoffman (1983; Hoffman & Richards, 1984) also extends these 2-D rules to 3-D, showing that surfaces can often be decomposed along extrema of negative principal curvature into physically meaningful parts. Note that in this scheme an object is described in terms of the intrinsic geometry of its surfaces and not in terms of volumetric primitives such as axes and cross-sections. Both schemes might well be incorporated in human vision, as these researchers demonstrate. Furthermore, the work pf Shepard and Metzler (1971) regarding the mental rotation of 3-D shapes strongly 30 STEVENS
2. VISUAL OBJECT PERCEPTION 31 suggests that the visual system is able to manipulate in an effectively continuous manner an internal representation of a seen object which captures salient part-whole relations (see Shepard, 1981 for discussion). The connection between this ability and objection perception is not at all clear, unfortunately. 3.5. Interpreting Contours in 3-D The visual system is quite adept at imposing 3-D interpretations on line drawings, the line-drawn Necker cube being a familiar example. Although there are infinitely many 3-D configurations thatmight have projected to any given 2-D image, we tend to see a specific configuration in 3-D (and its reversal)— clearly, we impose strong constraints on the 3-D interpretation. It is well established that in viewing such figures we prefer that interpreta- tion which favours regularities, symmetries, parallelisms, coplanarities, etc. Consequently, we tend to see parallelograms and trapezoids as slanted rectangles or squares, trihedral vertices as right trihedral vertices, and so forth (see, e.g., Perkins, 1982; Shepard, 1981; Stevens, 1983a). These tendencies might be indirect consequences of more general Praegnanz principles or in some cases they might directly reflect specific perceptual assumptions, (e.g., of perpendicularity). To amplify this point, the Praeg- nanz principle (Koffka, 1935) says that we tend to choose the simplest interpretation, particularly in cases of minimal stimuli, hence given a trapezoid in 2-D we prefer the 3-D interpretation which, in addition to preserving parallelism and symmetry, results in equal length lines and equal angles at the vertices (which implicitly results in 90° angles in 3-D). While there is considerable evidence for Praegnanz in our visual interpre- tations, it does not exclude the possibility that we also drive our interpretations by strongly constraining assumptions that more directly entail perpendicularity and other geometric properties (discussed later). As to whether such geometric assumptions can be rephrased in terms of Praegnanz is probably a pursuit of secondary importance. In contrast, of considerable importance from a computational point of view is determin- ing precisely what 3-D interpretation problems exist, what each entails, and what constraint is employed to solve each. It is another matter to study how these various constraints are integrated, that is, by what methods do simultaneous and potentially conflicting constraints force one interpreta- tion over another. Questions of how these constraints might be implemented within some perceptual process can be studied distinctly from the constraints themselves. It is there that issues of Praegnanz re-emerge (see, for example, the discussion on “hill-climbing” , i.e. the method of gradient ascent, discussed by Attneave, 1982 and various proposed
32 STEVENS computing architectures such as connectionist models, e.g., Ballard, Hin- ton, & Sejnowski, 1983; see also Humphreys & Quinlan, this volume, Chapter 3). In the following, we take the view that we can address the constraints that allow the interpretation of 3-D shape independently of how the constraints are implemented. The computer vision community examined the con- straint question in the limited domain of blocks (Waltz, 1975), showing how local configurations of vertices, while individually ambiguous, are mutually constraining so that a global 3-D interpretation can usually be derived, given assumptions of general position (ruling out accidental alignments) and a priori assumptions about the restricted domain (right trihedral vertices, planar facets, etc.). Note that the computer vision work is clearly related to the empirical observations by Hochberg (1982) con- cerning the accumulation of local geometric constraint when viewing a line drawing through a limited aperture. Consider now the 3-D interpretation of smooth curves. The approach has been to distinguish three physical categories of curve: (1) the bound- ing, or occluding, contours that comprise the silhouette boundary of a smooth shape (Marr, 1977,1982); (2) foreshortened texture contours such as the mottled pattern of light and shadow under a tree (Witkin, 1981); and (3) smooth contours that extend across a surface, such as seams, creases, wrinkles, and linear pigmentation markings (Stevens, 1981, 1986). These three, distinctly different categories of contour provide information about surface shape in different ways. Marr’s (1977) work on smooth silhouette boundaries concerns inferring the axis of an object and its shape about that axis, all on the basis of its silhouette shape in the image. Marr uses the term contour generator to refer to the 3-D locus of surface points that projects to the (silhouette) contour, and makes three geometrical assumptions about the contour generator that allow the 3-D interpretation: (1) each line of sight from the viewer to the object grazes the object’s surface at exactly one point; (2) that nearby points on the contour arise from nearby points on the contour generator; and (3) the contour generator lies wholly in a single plane. It would appear that these assumptions must be held a priori to make any concrete inferences about the shape of an object from its boundary. Interested readers are referred to Marr (1982) for continuation of this issue. What about the contours that lie interior to the silhouette of an object? Of these it is useful to distinguish between contours that follow or reflect the shape of the underlying surface (Fig. 2.1) from those whose shape is 5That is, the surface texture is random in that the contour curvature is not correlated with the tangent direction.
2. VISUAL OBJECT PERCEPTION 3 3 FIG. 2.1. Contours following or reflecting shape of underlying surface. random, but whose systematic foreshortening reflects the orientation of the approximately planar surface underneath (Fig. 2.2) (Stevens, 1981; Wit- kin, 1981). With reference to Fig. 2.2, Witkin (1981) has shown that the slant and tilt of a surface patch can be estimated from a measure of the foreshortening of texture contours (there is a lawful relationship between the magnitude of the contour curvature and the orientation of the tangent along the contour), provided the viewer assumes that the texture is isotropic5 and the surface is approximately planar under the contours. When these assumptions hold, the surface orientation predicted by that computational method corresponds rather well with the apparent surface orientation. Of course, if the surface texture is actually anisotropic in a manner that mimics foreshortening, the method would fail, giving an FIG. 2.2. Foreshortening of texture contours.
Random documents with unrelated content Scribd suggests to you:
The Project Gutenberg eBook of False Evidence
This ebook is for the use of anyone anywhere in the United States and most other parts of the world at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this ebook or online at www.gutenberg.org. If you are not located in the United States, you will have to check the laws of the country where you are located before using this eBook. Title: False Evidence Author: E. Phillips Oppenheim Illustrator: Maurice Grieffenhagen Release date: October 24, 2017 [eBook #55798] Most recently updated: October 23, 2024 Language: English Credits: Produced by Al Haines *** START OF THE PROJECT GUTENBERG EBOOK FALSE EVIDENCE ***
"They sprang after me, but started back with a quick exclamation, for they looked into the black muzzle of my father's revolver." (Chapter XXXVII.) FALSE EVIDENCE BY E. PHILLIPS OPPENHEIM Author of "Anne, the Adventuress," "The Traitors," "Conspirators," etc. WARD, LOCK & CO., LIMITED, LONDON, MELBOURNE AND TORONTO. 1911. This Book, written by the Author some years ago, is now issued in Library form for the first time.
CONTENTS CHAP. PROLOGUE I. MY APOLOGY II. THE FIRST CLOUD III. "THE BOY MUST BE TOLD" IV. "A MYSTERIOUS MEETING" V. "ON BOSSINGTON HEADLAND" VI. AN INTERRUPTED ADDRESS VII. "I AM TOLD" VIII. "MY VOW" IX. AN UNEXPECTED VISIT X. THE FIRST MOVE XI. COLONEL DEVEREUX'S LAND AGENT XII. AT DEVEREUX COURT XIII. COLONEL SIR FRANCIS DEVEREUX, BART. XIV. THE BEGINNING OF DANGER XV. A FIGHT FOR LIFE XVI. MY CONVALESCENCE XVII. A MOONLIGHT RIDE XVIII. A STRANGE INTERVIEW XIX. MARIAN SURPRISES ME XX. AMONGST THE BULRUSHES XXI. RUPERT DEVEREUX XXII. FACE TO FACE XXIII. IN THE PICTURE GALLERY XXIV. A MIDNIGHT VISITOR XXV. "COUSINS!" XXVI. I "GIVE WARNING" XXVII. SIR FRANCIS DEVEREUX'S APPEAL XXVIII. GOOD-BYE TO DEVEREUX COURT
XXIX. I AM TEMPTED XXX. LIAR AND COWARD XXXI. MY FATHER AND I XXXII. THE BRIGANDS' HOME XXXIII. AT PALERMO XXXIV. VISITORS FROM ROME XXXV. WE ENTERTAIN AT THE VILLA XXXVI. MR. BURTON LEIGH XXXVII. CUT DOWN XXXVIII. AN OMINOUS NOTE XXXIX. "MY FATHER'S RESOLUTION" XL. A HORRIBLE MISTAKE XLI. "TWO YEARS AFTER" XLII. A TRAITOROUS LOVE XLIII. EXPIATION XLIV. "HERO"
FALSE EVIDENCE PROLOGUE The last sally had been made and repulsed, the last shot fired; the fight was over, and victory remained with the white men. And yet, after all, was it a victory or a massacre? If you were a stay-at-home, and read the report from the telegrams in your club, or in the triumphant columns of the daily papers, especially those on the side of the Ministry, you would certainly have pronounced it the former. But if you had been there on the spot, and had seen the half-naked, ill-armed natives, with the fire of patriotism blazing in their eyes and leaping in their hearts—had seen them being shot down in rows by the merciless guns of the English batteries—another view of the matter might have presented itself to you. It might have occurred to you that these men were fighting on their own soil for their freedom and their country, and that the spirit which was blinding their eyes to the hopelessness of resistance, and urging them on to resist the stranger's progress with such passionate ineffectiveness, was after all, a natural and a poetic one. But, after all, this has nothing to do with my story. The battle was over, and it was morning. Far away in the east a dull red light had arisen from over the tops of the towering black mountains, and an angry sun was sullenly shining on the scene of carnage. It was a low hillside, once pleasant enough to look upon, but at that moment probably the most hideous sight which the whole universe could have shown. The silvery streams, which had trickled lazily down to the valley below, now ran thick and red with blood. The luxuriant shrubs and high waving ferns were trampled down and disfigured, and, most horrible sight of all, everywhere were strewn the copper-coloured forms of the beaten natives. There they lay apart and in heaps in all imaginable postures, and with all imaginable
expressions on their hard, battered faces. Some lay on their sides with their fingers locked around their spears, and the rigid frown and convulsed passion of an undying hatred branded on their numbed features. Others less brave had been shot in the back whilst flying from the death-dealing fire of the European guns, and lay stretched about in attitudes which in life would have been comical, but in death were grotesquely hideous; and over the sloping fields the misty clouds of smoke still lingered and curled upwards from the battered extinct shells which lay thick on the ground. High above the scene of devastation, on a rocky tableau at the summit of the range of hills, were pitched the tents of the victors. A little apart from these, conspicuous by the flag which floated above it, were the general's quarters; and underneath that sloping roof of canvas a strange scene was being enacted. Seated amongst a little group of the superior officers, with a heavy frown on his stern face, sat the general. Before him, at a little distance, with a soldier on either side, stood a tall, slight young man, in the uniform of an officer, but swordless. His smooth face, as yet beardless, was dyed with a deep flush, which might well be there, whether it proceeded from shame or indignation. For he was under arrest, and charged with a crime which, in a soldier, is heinous indeed—it was cowardice. It was a court-martial before which he stood arraigned, although a hastily improvised one. But soldiers have prompt ideas of justice, and General Luxton was a martinet in all matters of discipline. Disciplinarian though he was, however, he liked little the task which was now before him. He looked up from the papers, which were stretched out on the rickety little round table, with a sudden movement, and bent his frowning gaze upon the accused. The young man returned his gaze steadily, but the colour in his cheeks grew deeper. "Herbert Devereux, you stand accused of a crime which, in your profession, nothing can palliate or excuse. Have you anything to say for yourself?"
"There will be no need for me to say anything, sir," was the prompt reply. "It is true that I turned my back upon the enemy, but it was to face a greater danger. The man whose life I saved can disprove this cruel charge against me in a moment. I admit that, from your point of view, appearances are suspicious, but you have only to learn from my half-brother, Rupert Devereux, why I quitted my post, and what I effected by so doing, to absolve me at least from all suspicion of cowardice, however much I may be to blame as a matter of discipline." General Luxton appeared surprised, a little relieved. "I hope so," he said, not unkindly. "Roberts, send an orderly to Lieutenant Devereux's tent, and command his presence at once." The man withdrew, and there was a few minutes' delay. Then the entrance to the tent was lifted up, and a tall, dark young man, with thin but decided features, and flashing black eyes, stepped forward. He was handsome, after a certain type, but his expression was too lifeless and supercilious to be prepossessing. General Luxton looked up and nodded. "Lieutenant Devereux, your half-brother, who stands accused of cowardice in the face of the enemy, appeals to you to give evidence on his behalf. Let us hear what you saw of him during the recent fighting." Eagerly, and with a confident light in his fair young face, the prisoner turned towards the man to whom these words were addressed. But slowly and deliberately the latter turned his back upon his half-brother without noticing his glance of appeal, and with a scornful light in his eyes. There was a slight murmur, and an interchange of looks amongst the few who were present at this significant action. "I do not know, General Luxton," he said, slowly, "what the prisoner can expect me to say likely to benefit him. He can scarcely be so mad as to expect me to shield him in this matter on account of our relationship, or to preserve the honour of our name, and yet I do not see why else he should have appealed to me. I saw very little of the affair, and would rather not
have seen that. I was riding to you, sir, with a message from Colonel Elliott; and, as I passed trench 4, I saw the prisoner suddenly leave his company and run towards me. He passed several yards to the left, and as he seemed to be hurrying along aimlessly, I called to him. He made no answer, but ——" "LIAR!" The word seemed hurled out with such a passionate intensity that every one started. General Luxton looked up angrily. "Silence, sir! You will have an opportunity of saying what you have to say presently. Proceed, Devereux." "As I was saying," Rupert Devereux continued calmly, without appearing to have noticed the interruption, "he made no answer, but seemed to wish to avoid me. As the message with which I was entrusted was an important one, I rode on and left him hurrying towards the rear." With a sterner air even than he had at first assumed, General Luxton turned towards the unfortunate young man who stood before him. He was standing as though turned to stone, with wide-open eyes, staring at the man who had just spoken, attitude and expression alike bespeaking an overpowering bewilderment. "You are at liberty to ask the witness any questions," the General said, shortly. For a moment there was a dead silence. Then the words came pouring out from his quivering lips like a mountain torrent. "Rupert, what have you said? What does this mean? Good God, are you trying to ruin me? Did I not run to your assistance because you were beset by those three blackguards? Didn't I kill two of them and save your life? You can't have forgotten it! Why are you lying? Hilton saw it all, and so did Fenwick. Where are they? My God, this is horrible!"
The deep flush had gone from his cheeks, and left him pale as death. Great beads of perspiration stood out upon his forehead, and there was a wild look in his deep blue eyes. But the man to whom he made his passionate appeal kept his back turned and heeded not a word of it. Instead of answering he addressed the General. "General Luxton," Rupert said, calmly, "the accused, in denying the truth of my statement, mentions the names of two men whom he admits were witnesses of this lamentable occurrence. Might I suggest that they be called to give their version?" The General nodded assent, and the thing was done. But Hilton was the only one who answered the summons, and on reference to a list of the killed and wounded it was found that Fenwick was reported missing. "John Hilton, the accused has appealed to you to give evidence on his behalf. Let us hear what you saw of him during the recent fighting." The man, an ordinary-looking private, stepped forward and saluted. "I only saw him for a moment, sir," he said, slowly, and with a marked reluctance. "I was riding behind Lieutenant Devereux when I saw him leave his company and pass us a few yards to the left. It struck me that he looked very pale, and I thought that perhaps he was wounded." "He did not leave his company to come to your master's assistance, then?" "Certainly not, sir. We were not in any need of it. None of the enemy were near us." "Thank you. You can go, Hilton." The man saluted and went. There was a dead silence for a full minute. Then there came a passionate, hysterical cry from the prisoner—
"Liar! Liar! General Luxton, upon my honour, either my brother and this man are under some hallucination or they have entered into a conspiracy against me. Before God Almighty I swear that I only left my post because several of the enemy had crept down from the hill behind and had attacked my brother and his servant. I killed one of them, and the blood of the other is still on my sword. Why, Rupert, you know that you called out, 'Thanks, Herbert, you have saved my life.' Those were your very words!" The man appealed to shook his head slowly and as though with great reluctance. The sigh seemed to madden the prisoner, and he made a sudden movement forward as though to spring at him. "Oh, this is horrible!" he cried. "Where is Fenwick? He saw it all. Let him be called." General Luxton glanced again at the list before him and looked up. "You are unfortunate in your selections," he said, dryly. "The evidence of Hilton and your brother, to whom you appealed, only strengthens the case against you. Fenwick is missing. Herbert Devereux," he went on sternly, "the charge against you has been proved. I, myself, at a most critical moment, saw you desert your post when it was the centre of attack, and it fell to another's lot to lead your men on to the pursuit. The reasons which you have brought forward to account for your unwarrantable action have been clearly disposed of. You are most certainly guilty of a crime for which, amongst soldiers, there is no pardon. But you are young, and I cannot forget that you are the son of one of the most distinguished officers with whom it has been my good fortune to be associated. For his sake I am willing to make some allowance for you—on one condition you may retain your commission, and, I trust, retrieve this well-nigh fatal mistake in the future. To the crime of cowardice you have added the crime of lying; for that your account of the attack upon your half-brother and your rescue is a pure fabrication I cannot doubt. The peculiar curve in the defile behind trench 4 unfortunately hid you from the field of battle and prevents further evidence as to the occurrence which, you say, took place. But that your story is false no one can possibly doubt. The place has been carefully examined, and there are no dead bodies within a hundred yards. It seems, from your appeal to your half-brother, that you expected him to shield you at the expense of
his honour. This lie and false statement of yours you must retract if you hope for any mercy from me." There was a convulsive agony in the boy's white, strained face as he drew himself up, and looked half piteously, half indignantly at his judge. But when he tried to speak he could not, and there was a minute or two's dead silence whilst he was struggling to obtain the mastery over himself. All expected a confession, and General Luxton removed his eyes from the prisoner, and bent close over his papers, that none might read the compassion which was in his heart, and which was reflected in his face. The words came at last; and shrill and incoherent though they were, there was a ring of genuine dignity in them. "General Luxton, I have been guilty neither of cowardice nor falsehood. I swear before God, on the sword which my father himself put into my hands before I left England; by everything that is most holy to me I swear that my account of this awful occurrence is true. Ask the men of whom I was in command when I caught sight of—of him"—and he pointed with a trembling finger and a gesture than which nothing could have been more dramatic to his half-brother—"ask them whether I bore myself like a coward when those spears were whistling around us, or when we were fighting hand-to-hand after the first repulse. God knows that I did not. I left my post to encounter a greater danger still. Bitterly do I regret that I ever did so; but it is the only indiscretion of which I am guilty. I swear it." General Luxton raised his head, and what there had been of compassion in his face was either gone or effectually concealed. "You have sworn enough already," he said, sternly. "Herbert Devereux, I am bitterly disappointed in you. I was willing to spare your father the disgrace which I fear will kill him; but you cut away the ground from under my feet. You are most certainly proved guilty of gross cowardice in the face of the enemy found guilty, not upon the evidence of one man, but of two, and one of those your own relative. Circumstances, too, are strong against you, so are the probabilities. Most undeniably and conclusively you are found guilty; guilty of cowardice, guilty of falsehood. You will remain under arrest until I can find an opportunity of sending an escort with you to
the Hekla. Your commission is forfeited to the Queen, whose uniform you have disgraced." Never a sign of guilt in the prisoner's countenance. Proudly and indignantly he looked his General straight in the face, his cheeks red with a flush, which was not of shame, and the wild fury in his heart blazing out of his eyes. "It is not I who have disgraced the Queen's colours; but he—he who has fabricated and sworn to a false string of lies. Rupert, in your heart alone is the knowledge of why you have done this thing. But some day you shall tell me—or die." There was something intensely dramatic in the passionate bitterness which vibrated in the shrill boyish tone, and, as though moved by a common impulse, every one in the tent followed that threatening gesture. But the face of Rupert Devereux was little like the face of a guilty man. He looked somewhat agitated, and a good deal pained; but although he was the cynosure of all eyes, he turned never a shade the paler, nor flinched once from the passionate fire which was leaping from the eyes of the young prisoner. He seemed as though about to make some reply; but the General raised his hand. "Remove the prisoner." There was a sudden commotion, for, with a deep, despairing groan, and arms for a moment lifted high above his head, he had staggered backwards and sunk heavily to the ground in a dead swoon. What wonder! He was but a boy after all. * * * * * "Herbert! Why, Herbert! Good God! where did you spring from? Are you invalided?" The moonlight was streaming in through the high oriel windows of the long picture-gallery, glittering upon the armour and crossed weapons which hung upon the walls, and casting fantastic rays down the polished oak floor.
Colonel Sir Francis Devereux dropped the cigar which he had been peacefully smoking, and brought to a sudden halt his leisurely perambulation of this his favourite resort. Before him, with drooping head, with sunken cheeks, and with deep black rims under his eyes, stood his son Herbert, who, only a few months ago, had departed on his first campaign, a happy, careless young sub. Was it, indeed, his son, or was it a ghost that had stolen upon him out of the gloomy shadows of the vast gallery? "Invalided! Would to God that I was dead!" broke from the boy's quivering lips. "Father, I have brought disgrace upon you—disgrace upon our name." And he stretched out his hands towards the long line of pictured warriors, who seemed to be frowning down upon him from the wall. "Disgrace that you will never forgive, never pardon." Like a statue of stone the proud old soldier stood while he listened to his son's story. Then, with a half-smothered groan, he deliberately turned his back upon him. "Father," he pleaded, "listen to me. Before heaven I swear that I am innocent. Rupert lied. Why, I don't know, but he lied. I never felt fear." His father turned half round. "You have been put on your defence. General Luxton would never have found your father's son guilty of cowardice had there been room for doubt. The charge was proved against you in court-martial." "But, father, it was because they believed Rupert and his man. The only two other men who saw the struggle are dead." Colonel Devereux turned away and buried his face in his hands. "A Devereux guilty of cowardice!" he groaned. "My God! that it should have been my son!" Then with a sudden movement he turned round. His son had sunk upon his knees before him, and the moon was throwing a ghastly light upon his haggard, supplicating face.
"Out of my sight, and out of my heart for ever, Herbert Devereux!" cried his father, his tones vibrating with a passionate contempt. "You have brought disgrace upon a stainless name. Curse you for it, though you be a thousand times my son. You shall not sleep under this roof again. Begone! Change your name, I command you! Forget that you are a Devereux, as I most surely shall. Turn linen-draper, or man-milliner, or lawyer, what you will so that I never see or hear from you again. Begone, and curse you." Scathing and vibrating with scorn though the words were, they seemed to touch a chord in the boy's heart, not of humiliation, but of righteous anger. He sprang to his feet, and held himself for a moment as proudly as any of his armoured ancestors who looked down from the walls upon father and son. "I will go, then," he cried, firmly. "It is right that I should go. But, after all, it is false to say that I have disgraced your name. It is Rupert who has done this." He turned and walked steadily away, without a backward glance. Out of the swing doors on to the broad staircase, he passed along noble corridors, between rows of marble statues, down into the mighty dome-like hall, and out of the house which he had loved so well. And the servants, who would have pressed forward to welcome him, hung back in fear, for there was that in his face which they shrunk from looking upon. Out into the soft summer night he stepped, heedless of their wondering glances, and down the broad avenue he hurried, never pausing once to breathe in the balmy night wind, heavy with the odour of sweet-smelling flowers, or to listen to the nightingale singing in the low copse which bordered the gardens. Through a low iron gate he stepped into the park, and walked swiftly along, never glancing to the right or to the left at the strange shadows cast by the mighty oak-trees on the velvety turf, or at the startled deer, who sprung up on every side of him and bounded gracefully away, or at the rabbits who were scampering about all around in desperate alarm; once he had loved to watch and to listen to all these things; but now he felt only a burning desire to escape from them, and to find himself outside the confines of the home which he was leaving for ever. And not until he had reached the last paling, and had vaulted into the broad, white road, did his strength desert him.
Then, faint and weary, and heartsick, he sank down in a heap on the roadside, and prayed that he might die. * * * * * A cloudless summer morning, with the freshness of dawn still lingering in the air. A morning which seemed about to herald in one of Nature's perfect days, on which to be sad were a crime, and to have troubles absurd. Already the dreamy humming of bees was floating in the atmosphere, and the lark had given place to noisier, if less musical, songsters. It was a glorious morning. Over the low, iron gate of an old-fashioned garden a girl was leaning, her head resting lightly upon her hand, gazing across the pleasant meadows to the dark woods beyond, with a soft, far-away look in her grey eyes—for she was thinking of her lover. She was dressed in a blue print gown, which hung in simple folds around her straight, slim figure, and she had carelessly passed the long stalk of a full-blown red rose within her waistband. It was a very pleasant view that she was admiring; but any casual spectator would have declared that she was the most charming object in it. And there was a spectator, although not a casual one. Suddenly, like a ghost, the figure of her dreams stood before her. Pale, haggard, and dishevelled-looking, he seemed to have risen out of the very ground; and it was very little to be wondered at that, at first, she shrunk back alarmed. "Herbert! Herbert! can it really be you?" He never answered her; but, as the first surprise began to fade away, she moved forward, and would have thrown herself into his arms. But he stopped her. "Keep back, Marian," he cried, hoarsely; "keep away from me! I have come to bid you good-bye." A swift, sudden fear drove the colour from her cheeks, and chilled her through and through; but she faltered out an answer.
"Good-bye, Herbert! What do you mean? Oh, tell me what has happened, quick!" "The one thing worse than death, Marian—disgrace!" And then, with his face turned away, and his eyes resting wearily on the picturesque landscape, he told her his story. * * * * * The last word had left his quivering lips, and he stood as though in a dream. The worst was over. He had told his father, and he had told her. It seemed like the end of all things to him. Suddenly a pair of white arms were thrown around his neck, and a great red rose was crushed to pieces against his waistcoat. "Herbert! oh, Herbert! how dreadful! Don't look like that, you frighten me!" He was striving to free himself, but she would not let him go. "Dearest, you don't understand! This is ruin to me. My father has turned me from the house, commanded me to bear another name, disowned me. Be brave, Marian, for we must part. I am here only to tell you this, and to bid you farewell." Still she would not let him go. "You will do nothing of the sort, sir. I'll not be thrown over in that fashion," she said, struggling to smile through her tears. "And, Herbert, oh, Herbert! how ill you look! You've been out all night." He did not deny it, but again he strove to disengage himself. But she would have none of it. "Bertie, dearest," she spoke cheerfully, though her eyes were still swimming with tears, "you mustn't think that you're going to get rid of us in
this way. You've just got to come in to breakfast with me, and afterwards we'll tell Grannie all about it. Come along, sir, I insist." He braced himself up for resistance, but he had still to learn that against a woman's love a man's will can prevail nothing. At first he was firm, then wavering, and finally he was led in triumph across the smooth lawn and along the winding path to the French windows of the morning-room. But when he found himself face to face with the kind old lady who had loved him as her own son, and saw the tears trickle down her withered, apple-red cheeks as she listened to the tale which Marian poured out, he felt that he had passed the limits of self-endurance. For more than twenty-four hours he had neither eaten nor drunk, and he was sick at heart. Gradually Marian felt the arm, which she had drawn tightly through hers, grow heavier and heavier until at last as she finished her tale with a little tremulous burst of indignation, he sank back in the arm-chair, and slowly fainted. But through the mist which closed in upon him he saw nothing but kindly pitying faces bending over his, and heard Grannie's gentle whisper— "I believe you, Herbert," and more emphatic but none the less earnest were her words, whose sweet, tear-stained face, so close to his, was the last he saw when unconsciousness was closing in upon him. "So do I, Bertie, I hate Rupert," and sweeter than the most heart-stirring music were the faltering words she added— "And I love you better than ever. Oh, Grannie, Grannie, he has fainted!" CHAPTER I MY APOLOGY Fortune is the strangest mistress a man ever wooed. Who courts her she shuns, who deserves her she passes over, and on him who defies her and takes no pains to secure her she lavishes her favours. I am one of those to
whom she has shown herself most kind. Many years ago I vowed my life away to one purpose, and that partly an immoral one. It was a purpose which held my life. I swore to seek no end apart from it, and I put away from my thoughts all joys that were not included in its accomplishment. And yet, having kept my oath, I still possess in the prime of life everything which a man could wish for. I am rich, and well thought of amongst my fellows. I am married to the woman whom I love, and life is flowing on with me as calmly and peacefully as the murmuring waters of a woodland stream in the middle of summer. And, above all, my heart is at ease, for I have kept my vow. She is a strange mistress, indeed! Nothing have I sought or deserved of her, yet everything I have. Whilst he who was far above me in his deservings, and whose sufferings none save myself thoroughly understood, passed through a gloomy life, buffeted by every wind, stranded by every tide of fortune; misunderstood, wronged, falsely accused, and narrowly escaped remaining in men's minds only as a prototype of a passionate, unforgiving, Quixotic man. That the world may know him as he was, and form a better judgment as to his character, I have gathered together the threads of my life indissolubly connected with his, and have turned them inside out. I have never indulged myself with the feminine luxury of a diary, but with a surer progress than of pen over paper has the record of my strange life been written into my mind; and so I tell it just as it all comes back to me, not as a professed story-teller, with harmonious dates and regular evolution of plot, and neatly paged chapters, but in a bolder way, leaving much to be guessed at, and some things untold. If there be any of whom I have occasion to speak still amongst the living (my life has so contracted of late that many have passed out of its horizon), let them remember for what purpose I write, and for his sake forbear to complain. If the sword were the pen, then would mine be the pen of a ready writer, and I might be able to touch lightly on their shortcomings, and gild over the black spots on my own life. But enough of excuses. I take up my pen a blunt Englishman, an athlete rather than a scholar, to write a plain story which shall serve not as a eulogy, but as a justification of the man to whom many years of my life have been
ungrudgingly given. Let all those who may feel disposed to cavil at the disconnectedness of my loosely jointed story, remember this, and be silent. CHAPTER II THE FIRST CLOUD About a mile seaward from Porlock, separated from it by a narrow strip of the most luxuriant meadowland in Devonshire, lies the village of Bossington. Perhaps it were better called a hamlet, for at the time when I knew anything about it (which, let the tourist remember, is many years ago) it consisted but of six or seven cottages, a farmhouse, and a half-ruined old manor-house, for the privilege of living in which my father paid ten pounds a year, or some such trifling sum, to the neighbouring clergyman whose property it was. But what the place lacked in size was certainly atoned for—and more than atoned for—by the beauty of its situation. High above it, like a mighty protecting giant, rose Bossington Headland, covered always with a soft, springy turf, and glowing in midsummer with the brilliant colouring of rich purple heather and yellow gorse. Often have I stood on its highest point, and with my head bared to the strong fresh breeze, watched the sun rise over the Exmoor Hills and Dunkerry Beacon, and waited until it shed its first warm gleams on the white cottages and queer old church-tower of Porlock, which lay clustered together in picturesque irregularity at the head of the little bay. And almost as often have I gazed upon the same scene from the same spot by the less distinct but more harmonious light of the full harvest moon, and have wondered in which guise it seemed the fairest. Behind Bossington lay Allercombe Woods, great tree-covered hills sloping on one side down to the road which connected, and still connects, Porlock with Minehead and the outside world, and on the other, descending precipitously to the sea; so precipitously indeed that it seemed always a wonder to me how the thickly growing but stunted fir-trees could preserve
their shape and regularity. The descent from Bossington Headland into Porlock was by a steep winding path through Allercombe Woods, and many a time I have looked through the thin coating of green leaves upon the fields which stretched like a piece of patchwork below down to the sea, and wondered whether any other country in the world (I had never been out of Devonshire then) could be more beautiful than this. Within a stone's throw of where the blue sea of our English Bay of Naples rippled in on to the firm white sands, was the tumble-down old building in which we lived. What there had been of walls had long before our time been hidden by climbing plants and ivy, and in summer-time the place from a distance somewhat resembled a gigantic nosegay of cottage roses, jessamine, and other creeping flowers. There was but a small garden and no ground, for Bossington Headland rose precipitously close to the back of the house, and in front there was no space for any. A shed served as a stable for one or two Exmoor ponies, and also as a sleeping-place for the lanky, raw-boned Devonshire lad whom we kept to look after them. There were but few habitable rooms in our mansion, but they were sufficient, for our household was a small one. My father, mother, sister, myself, and a country servant comprised it. We never had a visitor, save occasionally the clergyman from Porlock. We never went anywhere. We knew no one, and at seventeen years of age an idea which had been developing in me for a long time, took to itself the tangible shape of words. "Father," I said to him one evening when we were sitting out upon our little strip of lawn together, he smoking, I envying him for being able to smoke, "do you know that I have never been out of Devonshire—never been further than Exeter even, and I am eighteen years old?" It was long before he answered me, and when, at last, he turned round and did so, I was distressed to see the look of deep anxiety in his worn, handsome face, and the troubled light in his clear eyes. "I know it, my boy," he said, pityingly. "I have been expecting this. You are weary of the country."
I stood up, with my hands in my pockets, and my back against the latticed wall of the house, gazing over the sparkling, dancing sea, to where, on the horizon, the stars seemed to stoop and meet it. Was I tired of this quiet home? I scarcely knew; country sports and country sights were dear to me, and I had no desire to leave them for ever. I thought of the fat trout in the Exford streams, and the huntsman's rallying call from "t'other side Dunkerry," and the wild birds that needed so much getting at and such quick firing, and of the deep-sea fishing, and the shooting up the coombes from Farmer Pulsford's boat, and of the delight of shipping on a hot summer's day and diving deep down into the cool bracing water. Why should I wish to leave all this? What should I be likely to find pleasanter in the world of which, as yet, I knew nothing? For a moment or two I hesitated —but it was only for a moment or two. The restlessness which had been growing up within me for years was built upon a solid foundation, and would not be silenced. "No, I'm not tired of the country, father," I answered, slowly. "I love it too much ever to be tired of it. But men don't generally live all their lives in one place, do they, without having any work or anything to do except enjoy themselves?" "And what should you like to be?" my father asked, quickly. I had long ago made up my mind upon that point, and was not slow to answer— "I should like to be a soldier," I declared, emphatically. I was very little prepared for the result of my words. A spasm of what seemed to be the most acute pain passed across my father's face, and he covered it for a moment with his hands. When he withdrew them he looked like a ghost, deathly pale in the golden moonlight, and when he spoke his voice trembled with emotion. "God forbid that you should wish it seriously!" he said, "for it is the one thing which you can never be!"
"Oh, Hugh, you do not mean it really; you do not wish to go away from us!" I turned round, for the voice, a soft and gentle one, was my mother's. She was standing in the open window with a fleecy white shawl around her head, and her eyes, the sweetest I ever saw, fixed appealingly upon me. I glanced from one to the other blankly, for my disappointment was great. Then, like a flash, a sudden conviction laid hold of me. There was some great and mysterious reason why we had lived so long apart from the world. CHAPTER III "THE BOY MUST BE TOLD" That was quite an eventful night in our quiet life. Whilst we three stood looking at one another half fearfully—I full of this strange, new idea which had just occurred to me—we heard the latch of our garden gate lifted, and Mr. Cox, the vicar of Porlock and my instructor in the classics, followed by no fewer than four large-limbed, broad-shouldered, Porlock men, entered. They made their way up the steep garden path, and my father, in no little surprise, rose to greet them. With Mr. Cox he shook hands and then glanced inquiringly at his followers, who, after touching their hats respectfully, stood in a row looking supremely uncomfortable, and each betraying a strong disposition to retire a little behind the others. Mr. Cox proceeded to explain matters. "You are pleased to look upon us as a deputation," he said, pleasantly, waving his hand towards the others, "of which I am the spokesman. We come from the Porlock Working Men's Conservative Club." My father bowed, and bidding me bring forward a garden seat, requested the deputation to be seated. Then he called into the house for Jane to bring out some jugs of cider and glasses, and a decided smile appeared on the
somewhat wooden faces of the deputation. I was vastly interested, and not a little curious. When the cider had been brought and distributed, and a raid made upon the tobacco jar, Mr. Cox proceeded with his explanation. "We have come to ask you a favour, Mr. Arbuthnot," he said. "We are going to hold a political meeting in the school-room at Porlock next week. A gentleman from Minehead is going to give us an address on the land question which promises to be very interesting, and Mr. Bowles here has kindly promised to say a few words." The end man on the seat here twirled his hat, and, being nudged by his neighbour, betrayed his personality by a broad grin. Finally, to relieve his modesty, he buried his face in the mug of cider which stood by his side. "The difficulty we are in is this," continued Mr. Cox; "we want a chairman. I have most unfortunately promised to be in Exeter on that day and shall not be able to return in time for the meeting, or else we would not have troubled you. But as I shall not be available, we thought that perhaps you might be induced to accept the office. That is what we have come to ask you." My father shook his head. "It is very kind of you to think of me," he said, hesitatingly, "but I fear that I must decline your offer. Politics have lost most of their interest for me —and—and, in short, I think I would rather not." "I hope you will reconsider that," Mr. Cox said, pleasantly. "It will be a very slight tax upon you after all. You need only say a very few words. Come, think it over again. We really are at our wit's end or we would not have troubled you. "There is Mr. Sothern," my father protested. "He is in bed ill. An attack of pleurisy, I think."
"Mr. Brown, then?" "A rank Radical." "Mr. Jephcote?" "Away." "Mr. Hetton?" "Gone to London for a week." "Mr. Smith, then?" "Will be at Exeter cattle fair." My father was silent for a moment or two. Then he suggested some more names, to each of which there was some objection. "You do seem to have been unfortunate," he declared, at last. "To tell you the truth, Mr. Cox," he added, thoughtfully, "I scarcely know what to say. I had made up my mind, for certain private reasons, never to have anything to do with public life in any shape or form." "This isn't a very formidable undertaking, is it?" Mr. Cox urged, smiling. "It isn't. But the principle is the same," my father answered. "However, leave it in this way if you like. Give me until to-morrow evening to think the matter over, and in the meantime see if you can't find some one else. I'm afraid I can't say more than that." The deputation thought that nothing could be fairer than this, and nothing more satisfactory except an unqualified assent. I think my father imagined that having promised so much they would take their departure. But nothing of the sort happened. Perhaps they found the cider too good, or perhaps they were tired after their day's work and the walk from Porlock. At any rate, there they sat for more than an hour, taking occasional gulps at their cider, and puffing incessantly at their blackened pipes with a stolid vacuous look on their honest faces, whilst my father and Mr. Cox talked a
little aside in a low tone. I fancied that I was the subject of their conversation, but though I strained my ears in the attempt to catch some part of it, I was unsuccessful. Once or twice the sound of my name reached me, but directly I leaned forward they dropped their voices, so that I could hear no more. I have always believed, however, that my father was asking advice from Mr. Cox concerning me, and that Mr. Cox was urging him to send me to the University. But I never knew for certain, for events were soon to occur which swept out of my mind all minor curiosity. At last Mr. Cox rose to go, and the deputation, with manifest reluctance, did the same. My father courteously accompanied them to the garden gate, and shook hands with them all, thanking them for their visit. When he returned there was a slight sparkle in his eyes, and an amused smile on his lips. So monotonous was our life, that even such an event as this was welcome, and I could tell from his manner that he was pleased at the request which had been made to him, and disposed to accept it. I determined to encourage him in it. "Governor," I remarked, leaning over the wall and watching the retreating forms of our visitors, "I hope we're not going to have many political deputations here, especially if they're all going to be as thirsty as this one was. Did you ever see such fellows for cider! We shan't have a drop left for the hot weather if you encourage this sort of thing. But you'll do what they want you to, won't you? I should! It'll be capital fun, and I'm sure you'd make a rattling speech. You're up on the land question, too. I heard you giving it to old Simpson the other morning." My father smiled, and stood by my side watching them make their way down the coombe. "I shall have to consult your mother about it," he said. "I almost think that I may venture it," he added, in a lower tone and thoughtfully, as though to himself. "Venture it! What could there be adventurous in it," I wondered, "to a well-read, scholarly man such as I knew him to be!" But I did not dare to ask.
Presently he turned to me with a much graver look in his face. "Hugh!" he said, "these people interrupted our conversation. There is something which I must say to you at once. I do not wish you to become a soldier. When you feel that you can stay here no longer, and that this country life is too quiet for you, you must choose some other profession. But a soldier you can never be." I was bitterly disappointed, and not a little curious, and an idea which had often occurred to me swept suddenly into my mind with renewed strength. "Father, may I ask you a question?" He hesitated, but did not forbid me. "I have heard it said down in the village—every one says that you must once have been a soldier. You walk and hold your head like one, and— father, what is the matter?" I broke off all at once, for his face had become like a dead man's, and he had sunk heavily on to the seat. I would have sprung to his side, but my mother was there before him. She had passed one arm around his neck, and with the other she motioned me to go into the house. "It isn't your fault, Hugh," she said, "but you mustn't ask your father questions; they distress him. Leave us now." I turned heavily away, and went up-stairs to my room. About an hour afterwards, when I pushed open my window before getting into bed, there stole into my room together with the sweet scent of jessamine and climbing roses the sound of subdued voices. "He must be told," I heard my father say solemnly. "God give me strength." Then the voices ceased for a while, but I still lingered, and presently they began again, but in a more cheerful key.
Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com

More Related Content

PDF
Visual Agnosia Second Edition 2nd Edition Martha J Farah
byzemajnikho
 
PDF
Imagery And Spatial Cognition Methods Models And Cognitive Assessment Tomaso ...
byharviseitzxe
 
PPTX
VISUAL COGNITION (ARIFA HUSSAIN).pptx
byArifaHussain10
 
PPSX
visual perceptual dysfunctions and assessment
byFarvardin Neuro-Cognitive Training Group
 
PDF
From Fragments To Objects Segmentation And Grouping In Vision 1st Thomas F Sh...
bykollolsonia
 
PDF
Visual Processing, Visual Perception & Visual-Motor Deficit.pdf
byOccupational Therapist
 
PPTX
Visual perceptual
byGundeep Singh
 
PDF
The Visual Brain In Action 2nd Edition A David Milner Melvyn A Goodale
bymiletoxainab98
 
Visual Agnosia Second Edition 2nd Edition Martha J Farah
byzemajnikho
 
Imagery And Spatial Cognition Methods Models And Cognitive Assessment Tomaso ...
byharviseitzxe
 
VISUAL COGNITION (ARIFA HUSSAIN).pptx
byArifaHussain10
 
visual perceptual dysfunctions and assessment
byFarvardin Neuro-Cognitive Training Group
 
From Fragments To Objects Segmentation And Grouping In Vision 1st Thomas F Sh...
bykollolsonia
 
Visual Processing, Visual Perception & Visual-Motor Deficit.pdf
byOccupational Therapist
 
Visual perceptual
byGundeep Singh
 
The Visual Brain In Action 2nd Edition A David Milner Melvyn A Goodale
bymiletoxainab98
 

Similar to Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys

DOCX
218 Perceiving visually presented objects recognition, aw.docx
bytamicawaysmith
 
PDF
Assessment_and_Intervention_of_Visual_Perception_and_Cognition3.pdf
byMaria Airam
 
PPTX
Eye Tracking Tests for Cognitive Impairment
byjetweedy
 
PDF
Object Recognition Attention And Action 1st Edition Ingo Rentschler
bydekerfdonova
 
PDF
Active Vision The Psychology of Looking and Seeing 1st Edition John M. Findlay
bywethalverko
 
PDF
Lecture3:Chapter5-Perception..Dr.Anna
byAHS_student
 
PDF
Visual Information Processing Perspectives On Fundamental Processes In Intell...
byromeuromps64
 
PDF
Cognitive Vision David E Irwin And Brian H Ross Eds
byituquvmnqp313
 
PDF
Perception
byDr.Ravi Ranjan Kumar
 
PPSX
Vision and agnosias
byFarvardin Neuro-Cognitive Training Group
 
PDF
Visual Perception Key Readings 1st Edition Steven Yantis
bycockswasiqi
 
PPT
12424498.ppt
byIbraHim Sartawi
 
PDF
Visual Impairment In Children Due To Damage To The Brain 1st Edition Gordon D...
bymiludyuntao94
 
PPTX
5.visual hearing impairment
byReza Parker, MD
 
PPT
Recognising Visual and Auditory Stimuli
byhilmie87
 
DOCX
· What does the Goodale and Humphrey (1998) article mean by the f
byLesleyWhitesidefv
 
PDF
Space Objects Minds And Brains Essays In Cognitive Psychology 1st Edition Lyn...
byiifqyjha1504
 
PDF
Seeing And Visualizing Its Not What You Think Zenon W Pylyshyn
bymoznupro
 
PDF
The Visual Neurosciences Chalupa Lm Werner Js Eds
bybuallipuchol
 
PPSX
Human vision
byFarvardin Neuro-Cognitive Training Group
 
218 Perceiving visually presented objects recognition, aw.docx
bytamicawaysmith
 
Assessment_and_Intervention_of_Visual_Perception_and_Cognition3.pdf
byMaria Airam
 
Eye Tracking Tests for Cognitive Impairment
byjetweedy
 
Object Recognition Attention And Action 1st Edition Ingo Rentschler
bydekerfdonova
 
Active Vision The Psychology of Looking and Seeing 1st Edition John M. Findlay
bywethalverko
 
Lecture3:Chapter5-Perception..Dr.Anna
byAHS_student
 
Visual Information Processing Perspectives On Fundamental Processes In Intell...
byromeuromps64
 
Cognitive Vision David E Irwin And Brian H Ross Eds
byituquvmnqp313
 
Perception
byDr.Ravi Ranjan Kumar
 
Vision and agnosias
byFarvardin Neuro-Cognitive Training Group
 
Visual Perception Key Readings 1st Edition Steven Yantis
bycockswasiqi
 
12424498.ppt
byIbraHim Sartawi
 
Visual Impairment In Children Due To Damage To The Brain 1st Edition Gordon D...
bymiludyuntao94
 
5.visual hearing impairment
byReza Parker, MD
 
Recognising Visual and Auditory Stimuli
byhilmie87
 
· What does the Goodale and Humphrey (1998) article mean by the f
byLesleyWhitesidefv
 
Space Objects Minds And Brains Essays In Cognitive Psychology 1st Edition Lyn...
byiifqyjha1504
 
Seeing And Visualizing Its Not What You Think Zenon W Pylyshyn
bymoznupro
 
The Visual Neurosciences Chalupa Lm Werner Js Eds
bybuallipuchol
 
Human vision
byFarvardin Neuro-Cognitive Training Group
 

Recently uploaded

PDF
Pharmaceutical Quality Assurance Unit 1 (BP606T)
byChetan Mali
 
PPTX
Q4_PPT MUSIC & ARTS 7 week 1-2, matatag curriculum
byZEN
 
PPTX
How to Configure Dispatch Management System in Odoo 18 Inventory
byCeline George
 
PDF
How "Raiders of the Lost Ark" and "Ordinary People" Employ Non-Verbal Acting
by6x5csns4pn
 
PPTX
Greengnorance Toolkit Module 4 Active Mobility and Transport
byKarl Donert
 
PDF
Beyond the Absurd: A Comparative Reading of Waiting for Godot and the Bhagava...
byKruti Vyas
 
PDF
Family and Marriage __ Sociology __ Jhasketan Kuanar __ NURSING VIBE ONLY .pdf
byJK NURSING VIBES ONLY JHASKETAN KUANAR
 
PPTX
Plant fibres used as surgical dressings & Sutures – Surgical Catgut and Ligat...
bySai Meer College of Pharmacy
 
PDF
Judgement Regarding Land Acquisition Act not Applicable to Encroachers.pdf
byssuser9d8e4e
 
PDF
Holm Community Heritage at St Nicholas Kirk - 2025 AGM Minutes (30.04.2025)
byAntoine Pietri
 
PPTX
Fluorimetric Analysis- Theory, Instrumentation and Application
bySayali Powar
 
PPTX
MEMORY &FORGETTING. Shilpa Hotakar.Psychology pptx
bySHILPA HOTAKAR
 
PDF
Pratishta Educational Society., Courses & Opportunities
byGanapathiVankudoth
 
PDF
Bones by Sadu Kassam (play) story ppt pdf
byRonnieMaeBorres
 
PDF
Chapter 05 Drugs Acting on the Central Nervous System: Anti-Depressant Drugs
bySandeshSul
 
PDF
Orlando by Virginia Woolf: The Granite and The Rainbow
byAdityarajsinh Gohil
 
PPTX
Greengnorance Toolkit Module 5 Waste Management
byKarl Donert
 
PDF
Unit3 Phases and process of curriculum development.pdf
byziaoria016
 
PPTX
How to Manage Checkout Policy & Customer Portal in Odoo 18 Website
byCeline George
 
PPTX
Math 8 Quarter 4 Week 3 PRIMARY DATA.pptx
byshahanieabbat3
 
Pharmaceutical Quality Assurance Unit 1 (BP606T)
byChetan Mali
 
Q4_PPT MUSIC & ARTS 7 week 1-2, matatag curriculum
byZEN
 
How to Configure Dispatch Management System in Odoo 18 Inventory
byCeline George
 
How "Raiders of the Lost Ark" and "Ordinary People" Employ Non-Verbal Acting
by6x5csns4pn
 
Greengnorance Toolkit Module 4 Active Mobility and Transport
byKarl Donert
 
Beyond the Absurd: A Comparative Reading of Waiting for Godot and the Bhagava...
byKruti Vyas
 
Family and Marriage __ Sociology __ Jhasketan Kuanar __ NURSING VIBE ONLY .pdf
byJK NURSING VIBES ONLY JHASKETAN KUANAR
 
Plant fibres used as surgical dressings & Sutures – Surgical Catgut and Ligat...
bySai Meer College of Pharmacy
 
Judgement Regarding Land Acquisition Act not Applicable to Encroachers.pdf
byssuser9d8e4e
 
Holm Community Heritage at St Nicholas Kirk - 2025 AGM Minutes (30.04.2025)
byAntoine Pietri
 
Fluorimetric Analysis- Theory, Instrumentation and Application
bySayali Powar
 
MEMORY &FORGETTING. Shilpa Hotakar.Psychology pptx
bySHILPA HOTAKAR
 
Pratishta Educational Society., Courses & Opportunities
byGanapathiVankudoth
 
Bones by Sadu Kassam (play) story ppt pdf
byRonnieMaeBorres
 
Chapter 05 Drugs Acting on the Central Nervous System: Anti-Depressant Drugs
bySandeshSul
 
Orlando by Virginia Woolf: The Granite and The Rainbow
byAdityarajsinh Gohil
 
Greengnorance Toolkit Module 5 Waste Management
byKarl Donert
 
Unit3 Phases and process of curriculum development.pdf
byziaoria016
 
How to Manage Checkout Policy & Customer Portal in Odoo 18 Website
byCeline George
 
Math 8 Quarter 4 Week 3 PRIMARY DATA.pptx
byshahanieabbat3
 

Visual Object Processing A Cognitive Neuropsychological Approach Glyn W Humphreys

  • 1.
    Visual Object ProcessingA Cognitive Neuropsychological Approach Glyn W Humphreys download https://ebookbell.com/product/visual-object-processing-a- cognitive-neuropsychological-approach-glyn-w-humphreys-52461610 Explore and download more ebooks at ebookbell.com
  • 2.
    Here are somerecommended products that we believe you will be interested in. You can click the link to download. Visual Object Tracking Using Deep Learning Ashish Kumar https://ebookbell.com/product/visual-object-tracking-using-deep- learning-ashish-kumar-52598712 Visual Object Tracking From Correlation Filter To Deep Learning 1st Ed 2021 Weiwei Xing https://ebookbell.com/product/visual-object-tracking-from-correlation- filter-to-deep-learning-1st-ed-2021-weiwei-xing-36380984 Visual Object Recognition 1st Edition Grauman K Leibe B https://ebookbell.com/product/visual-object-recognition-1st-edition- grauman-k-leibe-b-4584652 The Visual Object Of Desire In Late Medieval England Sarah Stanbury https://ebookbell.com/product/the-visual-object-of-desire-in-late- medieval-england-sarah-stanbury-51968168
  • 3.
    Machine Learning ChallengesEvaluating Predictive Uncertainty Visual Object Classification And Recognising Tectual Entailment First Pascal Machine Learning Challenges Workshop Mlcw 2005 Southampton Uk April 1113 2005 Revised Selected Papers 1st Edition Joaquin Quionerocandela Auth https://ebookbell.com/product/machine-learning-challenges-evaluating- predictive-uncertainty-visual-object-classification-and-recognising- tectual-entailment-first-pascal-machine-learning-challenges-workshop- mlcw-2005-southampton-uk-april-1113-2005-revised-selected-papers-1st- edition-joaquin-quionerocandela-auth-4239698 Objectoriented Programming With Visual Basicnet Michael Mcmillan https://ebookbell.com/product/objectoriented-programming-with-visual- basicnet-michael-mcmillan-2004060 Beginning Object Oriented Programming Using Visual Basic 2005 From Novie To Professiona Dan Clark https://ebookbell.com/product/beginning-object-oriented-programming- using-visual-basic-2005-from-novie-to-professiona-dan-clark-4112518 Visual Perception And Robotic Manipulation 3d Object Recognition Tracking And Handeye Coordination 1st Edition Geoffrey Taylor https://ebookbell.com/product/visual-perception-and-robotic- manipulation-3d-object-recognition-tracking-and-handeye- coordination-1st-edition-geoffrey-taylor-983282 Visual Saliency From Pixellevel To Objectlevel Analysis 1st Ed Jianming Zhang https://ebookbell.com/product/visual-saliency-from-pixellevel-to- objectlevel-analysis-1st-ed-jianming-zhang-9964488
  • 6.
  • 8.
    VISUAL OBJECT PROCESSING ACognitive Neuropsychological Approach Edited by GLYN W. HUMPHREYS AND M. JANE RIDDOCH
  • 9.
    First published in1987 by Lawrence Erlbaum Associates Ltd This edition first published in 2017 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN and by Routledge 711 Third Avenue, New York, NY 10017 Routledge is an imprint o f the Taylor & Francis Group, an informa business © 1987 by Lawrence Erlbaum Associates Ltd All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-1-138-68824-7 (Set) ISBN: 978-1-315-22895-2 (Set) (ebk) ISBN: 978-1-138-20973-2 (Volume 15) (hbk) ISBN: 978-1-138-20976-3 (Volume 15) (pbk) ISBN: 978-1-315-45685-0 (Volume 15) (ebk) Publisher’s Note The publisher has gone to great lengths to ensure the quality of this reprint but points out that some imperfections in the original copies may be apparent. Disclaimer The publisher has made every effort to trace copyright holders and would welcome correspondence from those they have been unable to trace.
  • 10.
    Visual Object Processing: ACognitive Neuropsychological Approach Edited by Glyn W. Humphreys Department of Psychology, Birkbeck College,U.K. and M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, U.K.
  • 11.
    Copyright © 1987by Lawrence Erlbaum Associates Ltd All rights reserved. No part of this book may be reproduced in any form, by photostat, microform, retrieval system, or any other means without the prior written permission of the publisher. Lawrence Erlbaum Associates Ltd., Publishers 27 Palmeira Mansions Church Road Hove East Sussex, BN3 2FA U.K. British Library Cataloguing in Publication Data Visual object processing : a cognitive neuropsychological account. 1. Visual perception I. Humphreys, Glyn W. II. Riddoch, M. Jane 152.1/4 BF241 ISBN 0-86377-045-2 Typeset by Spire Print Services Ltd., Salisbury, Wilts Printed and bound by A. Wheaton & Co. Ltd., Exeter
  • 12.
    Contents List of Contributorsix Preface xi 1. Introduction: Cognitive Neuropsychology and Visual Object Processing 1 Glyn W. Humphreys and M. lane Riddoch 1. Introduction 1 2. Cognitive Neuropsychology and Visual Object Processing 2 3. The Book 11 2. Visual Object Perception from a Computational Perspective 17 Kent A. Stevens 1. Introduction 17 2. Modularity in Vision 19 3. Survey of Computational Problems 21 4. Concerning Visual Defects 35 5. Conclusion 39 3. Normal and Pathological Processes in Visual Object Constancy 43 Glyn W. Humphreys and Philip T. Quinlan 1. Introduction 43 2. The Problem of Visual Object Constancy 44 3. Approaches to Object Constancy 48 4. Animal Work 82 5. Neurological Impairments of Human Vision 88 6. Some Conclusions 95 V
  • 13.
    4. Picture Naming107 M. lane Riddoch and Glyn W. Humphreys 1. Introduction 107 2. Studies of Picture Naming in Normal Subjects 108 3. Data from Brain Damaged Subjects 127 4. Conclusions 139 5. Information Processing and Laterality Effects for Object and Face Perception 145 lustine Sergent 1. Introduction 145 2. Information Processing 148 3. Some Properties of the Visual System and their Implications 153 4. Hemispheric Processing Assymetry 162 5. Conclusion 167 6. The Clinical Spectrum and Localisation of Visual Agnosia 175 Andrew Kertesz 1. Introduction 175 2. Varieties of Visual Agnosia 176 3. Agnosia and Cortical Blindness 182 4. Associated Deficits 183 5. The Lesions Producing Visual Agnosia 189 6. The Functions of Visual Recognition 191 7. Apperceptive Agnosia: The Specification and Description of Constructs 197 lohn Campion 1. Introduction 197 2. A Neuropsychological Framework for Vision 198 3. Sensory Loss Theory and Agnosia 200 4. A Case of Apperceptive Agnosia 206 5. Two Other Patients 221 6. Conclusions 227 8. Object Concepts and Object Names: Some Deductions from Acquired Disorders of Word Processing 233 Elaine Funnell '~ 1. Modularity and the Division of Labour 233 2. Conceptual Representations of Objects 236 3. Levels of Object Recognition 239 4. Semantic Processing and Imageable/Concrete Word Concepts 243 5. Shared Conceptual Information for Object Names and Objects 253 9. Dementia and Visual Agnosia 261 Oscar S. M. Marin 1. Dementia: Definition and Nosology 261 2. Perceptual Disorders in Dementia 264 V¡ CONTENTS
  • 14.
    CONTENTS 3. Dementias asa Model for Neuropsychological Research 4. The Nature of Perceptual Disorders in Dementia 268 5. The Spectrum of Visual Agnosic Syndromes 275 10. The Fractionation of Visual Agnosia 281 Glyn W. Humphreys and M. Jane Riddoch 1. Introduction 281 2. Apperceptive Agnosia 282 3. Associative Agnosia 292 4. Classification and Lesion Sites 294 5. Re-classifying Visual Agnosia 299 6. Conclusions 304 Author Index 307 265 Subject Index 315 vii
  • 16.
    List of Contributors JohnCampion Behavioural Science Division, Admiralty Research Establishment, Queens Road, Teddington, Middlesex, U.K. Elaine Funnell Cognitive Neuropsychology Research Group, Department of Psychology, University of London, Malet Street, London WC1E 7HX, U.K. Glyn W. Humphreys Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street, London WC1E7HX, U.K. Andrew Kertesz Clinical Neurology Department; St. Joseph's Hospital, 268 Grosvenor Street, London, Ontario N6A 4V2, Canada Oscar S. M. Marin Laboratory of Cognitive Neuropsychology, Department of Neurology, Good Samaritan Hospital and Medical Center, Portland, Oregon 97210, U.S.A. Philip T. Quinlan Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street; London WC1E 7HX, U.K. M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, Romford Road, Stratford, London E15 4LZ, U.K. Justine Sergent Department of Neurology, McGill University, Montreal Neurological Institute, 3801 University Street, Montreal PQ, H3A 2B4, Canada Kent A. Stevens Department of Computer and Information Science, University of Oregon, Eugene, Oregon 97403, U.S.A. V.O.P.—A* ix
  • 18.
    Preface This book hasits origins in a meeting of the European International Neuropsychology Society (INS) at Aachen in June 1984. We had been asked to organise a symposium on visual agnosia at the conference. Our idea was to use the symposium to try to bring together workers from various disciplines in order to throw new light on the old problem of how visual object processing occurs, and on how it can be disturbed by damage to the brain. We invited a number of contributors to the book to give talks. The meeting seemed to go well, and there was interest expressed in the idea of using converging approaches to understanding object processing. Some time later, the thought occurred to us that a book covering these converging approaches might also serve some purpose, and this is the final product. In the period since the INS conference, work on all aspects of visual object processing has advanced considerably, and undoubtedly it will continue to do following the publication of this book. Our hope is to provide a state-of-the-art summary of work on the topic, and also to convey something of the excitement of working in a field subject to rapid development. The final editing of the book has been helped by many people: all the “ support staff” at Lawrence Erlbaum, the careful (though anonymous reviewers), and the Cognitive Neuropsychology Research Group at Birkbeck College, especially Philip Quinlan and Cathy Price. To all these people, our thanks. We are also grateful to Alan Cowey, Peter Dodwell, Geoff Hinton, Stephen Lupker, Stephen Palmer, and W. H. Freeman for xi
  • 19.
    x ii PREFACE permissionto reproduce figures. All the photographs were taken by Freddie Elliot, and Daphne Ring and Sue Godfrey provided much-needed last-minute typing. Our final thanks go to Iain, Alec, and Kate, for putting up with our working in the evenings!
  • 20.
    To Glyn andDorothy
  • 22.
    1 Introduction: Cognitive Neuropsychology and VisualObject Processing Glyn W. Humphreys Cognitive Neuropsychology Research Group, Department of Psychology, Birkbeck College, University of London, Malet Street, London WC1E 7HX, U.K. M. Jane Riddoch Department of Paramedical Sciences, North East London Polytechnic, Romford Road, Stratford, London E15 4LZ, U.K. The development of an efficient visual system has clearly been important for our survival. The ability to represent the position of objects with respect to our body position allows us to negotiate the world efficiently, so that we can avoid obstacles and reach appropriately for those objects we wish to grasp. The ability to recognise that an object has the same structure when seen from different viewpoints, with different lighting, etc. enables us to assign it the same response under a variety of contrasting viewing conditions. The ability to associate the viewed object with our memory facilitates such diverse activities as choosing food, taking early evasive action from predators, and selecting group members from amongst a crowd. Humans are also able to assign agreed names to objects, and so can communicate information about objects in an efficient manner. All of these different abilities reflect aspects of visual object processing. Normally, visual object processing is so effortless and (seemingly) immediate that it is difficult to conceptualise the complexity of the proces- ses involved. However, some inkling of this complexity can be gained when one encounters patients who, following brain damage, have selective dif- ficulties in such apparently simple abilities as matching objects presented from different viewpoints, knowing how to use a common object or assigning a name to it. Such cases suggest that visual object processing involves a set of rather independent abilities, each of which can be selectively affected by brain damage. 1. INTRODUCTION 1
  • 23.
    2 HUMPHREYS &RIDDOCH In this book, we have attempted to bring together work by researchers concerned with the functional and neurological mechanisms underlying visual object processing, and the ways in which such mechanisms can be neurologically impaired. We have termed it a “ Cognitive Neuropsycholog- ical” approach, because we believe it tries to relate evidence from neurological impairments of visual object processing to models of normal performance in a new and important way. Two broad aims are apparent. One is to test models of normal performance by evaluating how well the models account for the patterns of impairment and preservation of abilities that can occur following brain damage. The other is to use models of normal performance to further our understanding of acquired disorders of visual object processing. These aims distinguish the approach from neuro- psychological work whose primary aim is to relate acquired deficits to the sites of damage, and from work in the field of cognitive psychology which attempts only to develop models of normal performance (see also Coltheart, 1984, 1985). 2. COGNITIVE NEUROPSYCHOLOGY AND VISUAL OBJECT PROCESSING The characteristics of acognitive neuropsychological approach to behaviour are only currently developing and, to some extent, they will be shaped by the nature of the particular field of study. In at least some areas, progress has been limited by the lack of well-specified information processing models (see Coltheart, 1985). It seems clear that we will remain unable to explain or to predict the various ways in which a given ability may break down if we have little idea of the component processes involved. Recent developments in the field of visual object processing, however, provide grounds for optimism in this respect. In particular, the work on computer vision by David Marr and his associates (Marr, 1976, 1980, 1982; Marr & Nishihara, 1978) has produced a more developed informa- tion processing account of visual object processing than was hitherto available. Marr suggested that visual object processing could be considered in terms of a series of processing stages, each of which produces a more elaborated representation of a visually presented object, making different types of information explicit (see, for instance, Chapter 3 by Humphreys & Quinlan). Each of these processing stages operates as a separate module, which makes information available for subsequent stages but which is to a large extent functionally independent of those stages. Marr also outlined certain procedures which may determine how information at each stage is represented and transformed to provide the input for subse- quent processing stages. Perhaps by considering the types of representation mediating visual object processing, and the procedures which create or
  • 24.
    1. INTRODUCTION 3 enableaccess to these representations to occur, we may gain some insights into the disorders of object processing which are produced by brain damage. Also, instances of selective problems following brain damage may provide constraints on the models of normal performance. These, then, are the general aims of a cognitive neuropsychology approach to visual object processing. However, the application of this approach to vision has not proved so straightforward. There are probably several reasons for this. 2.1. Applying Theories from Computer Vision to Humans To begin with, there is an initial issue concerning the optimal way to “frame” the problems involved in visual object processing. As noted earlier, some of the most explicit models of the component processes involved are derived from the field of computer vision. Typically, such models are not aimed at accounting for how processing is carried out by humans. It follows that data from either normal observers or from subjects with acquired disorders of vision cannot be used to evaluate such models in their own right; the data may only be used to evaluate the models as accounts of human vision. The question arises as to which models are then appropriate. For instance, an axiom of computational approaches to information processing is that theories may be expressed at different levels of abstraction (see Stevens, Chapter 2). At a very high level of abstraction (usually termed a computational level), theories may be concerned only with what the requirements of the system may be. Thus Marr (1979) states that “ vision (requires) the construction of efficient symbolic descriptions from images of the world” . Descriptions may be constructed which make available information which is unique to the object seen from a given viewpoint (i.e., some form of viewpoint dependent description), enabling us to manipu- late the object in space. Descriptions may also be constructed which enable the object to address stored information irrespective of the viewpoint (i.e., some form of abstract object description), so that recognition can be achieved across different views (see Sutherland, 1979). According to this approach, vision requires the construction of each of the above kinds of description. Another level of theory would be concerned with how these descriptions are derived (the level of the algorithm), and yet another with how any processes are implemented in hardware terms (the level of the hardware; see Marr, 1982; Stevens, Chapter 2). We need to be clear about which level of theory is being addressed by any piece of behavioural evidence. Evidence that visual object processing in humans is highly constrained by viewpoint (see Humphreys & Quinlan, Chapter 3) would
  • 25.
    be pertinent tocomputational-level theories of the kind outlined by Marr (1982); evidence concerning the kinds of procedures used in constructing different types of visual description, however, would be pertinent to theories at the level of algorithm, not to computational level theories. The rejection of a theory as applied to human vision at the level of algorithm (e.g., in terms of the type of procedure used to generate a particular visual description) may not entail the rejection of the theory at a computational level (in the sense that that description may still mediate performance). Thus, the kinds of theory we accept will be dependent on the level of description to which any evidence applies. 2.2. Independence between Levels of Description The argument that there exist separate levels of description of any complex information processing problem makes sense in computer science, where the constraints on one description of a task (e.g., in terms of a flow-chart or a description of the algorithm) may be quite independent of those at another (such as hardware). It is not clear that this is true of the brain. Details concerning the speed of neuronal transmission, the specialisation of different visual areas, the prior tuning of cells within any visual area, etc., may be crucial in determining the manner in which different processes are carried out. If this suggestion is correct, then we may not be able to apply a theory at one level of description without considering its implementation at other levels in some detail. In cognitive neuropsychological approaches to other areas, particularly to the study of language and reading, evidence concerning the nature of the hardware has been effectively ignored (e.g. Coltheart, 1985; Mehler, Morton & Jusczyk, 1984). One reason for this selective view is that our knowledge of the brain structures mediating tasks such as reading has been insufficient to constrain theories concerned with reading mechanisms. This state of relative ignorance contrasts with that in the field of vision, where we now have quite detailed accounts of the neural systems involved, of their preferences for different types of visual input, and of their mapping (e.g. Cowey, 1985). It is possible that our developing knowledge of the neural hardware of vision will help determine theories of visual processing mechanisms. To illustrate this point, consider evidence concerning the selective preferences of cells in different visual areas of the cortex for different properties of visual stimuli. Recent work suggests that at least some cells in area V4 of the monkey cortex (see Fig. 11, Humphreys & Quinlan, Chapter 3) exhibit preferences for specific colours even when the wavelength of the illumination on a surface changes (i.e., they exhibit colour constancy; Zeki, 1980, 1983). Such cells might well be involved in colour identification (cf. Desimone, Schein, Moran, & Ungerleider, 1985; 4 HUMPHREYS & RIDDOCH
  • 26.
    Land, 1977). Theoriesconcerned with how we identify, say, a letter in a particular colour may need to take account of the possibility that the neural structures subserving colour identification may differ from those finally involved in letter identification. For instance, such theories may need to posit that colour and form identification are functionally independent (cf. Ballard, Hinton, & Sejnowski, 1983; Treisman & Gelade, 1980). It would follow that in order to identify a particular letter in a particular colour (e.g., against a background consisting of the same letter in a different colour and different letters in the target colour), some special mechanism may be required to co-ordinate information from the different visual areas, such as linking by common retinotopic locations (cf. Quinlan & Humphreys, 1987) or by the serial application of focal attention (Treisman & Souther, 1985). The process of interaction between the different levels of description need not be one way, however. Higher-level accounts of the mechanisms involved in visual object processing may also constrain theories at the hardware level. For instance, there is some evidence to suggest that the detection of boundaries between different visual textures can be deter- mined by activity in sets of detectors responding to simple local, properties of stimuli, such as colour and line orientation (cf. Beck, 1967, 1982; Treisman & Gelade, 1980). These properties correspond to discrimina- tions which may be expected from the activity of colour or orientation- tuned cells in the visual cortex (cf. Hubei & Weisel, 1968). Other behavioural evidence indicates that texture boundaries may also be distin- guished on the basis of more complex visual information, such as “ elon- gated blobs” and the number of line terminators (Julesz, 1980, 1981), or (possibly) closure (Treisman & Paterson, 1984). Given the argument that texture boundaries are signalled by specific hardware units tuned to the discriminating properties, it should follow that there exist detectors for the above complex types of information (see Julesz, 1980, 1986). Inspection of those theories currently shaping cognitive neuro- psychological work on vision indicates a fairly wide range of approaches to issues concerning the relations between different levels of theoretical description (although these differences are rarely made explicit). For example, Marr’s (1982) account of the processes leading up to the genera- tion of a three-dimensional (3-D) representation of a shape assumes that these processes may be described independently of their neural implemen- tation. Other processing accounts seem to be more intimately linked with the known properties of neural hardware. Most “ connectionist” theories of visual object processing at least purport to take seriously implications for the design of a system where complex knowledge is expressed via activity in a set of simple (neurone-like) computational units (see Hinton & Ander- son, 1981; Humphreys & Quinlan, this volume; McClelland & Rumelhart, 1. INTRODUCTION 5
  • 27.
    1986). Some connectionisttheories not only allocate separate processing units for specific visual properties of stimuli (such as lines at particular orientations and positions; see Fig. 8, Chapter 3, Humphreys & Quinlan, this volume), they also group units corresponding to properties within a given dimension (such as colour or size) to form a “ parameter space” , where the value(s) of a stimulus along that dimension is specified (cf. Ballard et al., 1983; Feldman & Ballard, 1982). Such parameter spaces may be considered analogous to areas of the visual cortex selectively tuned to different visual dimensions. What are the implicationsof these different approaches for the aims of cognitive neuropsychology we outlined earlier? First, consider how we may use evidence drawn from brain damaged patients to assess the different models. Models which maintain a clear independence between descriptions of processing mechanisms and hardware descriptions can only be addres- sed by information concerning the functional deficits of patients (i.e., by the type of behavioural deficit exhibited and by its relation to other component abilities). This position is akin to that typical of cognitive neuropsychological approaches in other areas, such as reading mentioned earlier. Models in which processing mechanisms are tied to the properties and locations of specific cortical areas may similarly be addressed by functional evidence. However, in addition, such models are also open to evaluation by evidence concerning the site and extent of damage. For instance, consider the possibility that a patient continues to exhibit colour constancy following selective ablation of area V4. This result would be difficult to accommodate for a theory which posited a colour-processing system explicitly modelled on our knowledge of the properties of cells in V4, and which maintained that colour-constancy could only be achieved using that colour-processing system. A theory positing a colour-processing system which does not have a one-to-one relationship with area V4 would not, of course, be similarly damaged. Accounts of brain damaged patients in terms of these different approaches will also differ. Models assuming independent algorithm-level and hardware-level descriptions account for behavioural deficits in patients in terms of patterns of dysfunction in (abstracted) processing components. Models assuming some explicit relations between the levels of description may additionally take on board information about the nature of the brain damage. Again, this may be illustrated by reference to theories positing that a given ability reflects patterns of activation across representations tuned to specific properties of stimuli (the connectionist accounts discussed by Humphreys & Quinlan, Chapter 3; the cascade model discussed by Riddoch & Humphreys, Chapter 4). Such theories may (eventually) offer different accounts of a particular deficit according to whether the deficit is due to a single selective lesion, due to a pattern of disseminated lesions (cf. 6 HUMPHREYS & RIDDOCH
  • 28.
    1. INTRODUCTION 7 Campion,Chapter 7) or due to a reduced density of connections within a neuronal network (cf. Marin, Chapter 9). Taken to their furthest conclu- sion, such theories may even return to the approach of classifying patients according to the nature (i.e. the type, extent, and site) of their brain damage, at least where our knowledge of the nature of the brain damage and its relation to behaviour is sufficiently precise. Campion, Chapter 7, argues for this approach to classifying patients with visual processing deficits following carbon monoxide poisoning. Whichever approach is adopted, we believe it important to realise that models can differ both in terms of their processing mechanisms and in terms of their assumptions concerning how the hypothesised processing mechanisms relate to the hardware of the system. The kinds of inferences we make will depend on the assumed processing— hardware relations. 2.3. Processing Modules The assumption that information processing is modular, in the sense that it is thought to comprise of a set of functionally independent processes, is one that is clearly of fundamental importance to cognitive neuropsychological approaches to behaviour (see Coltheart,1985). Such an assumption is at the heart of accounts of brain damaged patients in terms of a pattern of intact and impaired component abilities (see Funnell, Chapter 8). A principle of modularity has also been espoused by workers in computer vision (Marr, 1976; Stevens, Chapter 2). In the latter context, workers have suggested that breaking a large computation up into a series of separate modules is useful because a small change to one process will not have consequences on others. By specifying a complex behaviour in terms of a series of separate processing modules, direct links between computer vision theories and theories of human behaviour seem to be facilitated. However, it is relevant to point out that concepts of modularity can differ. One view, outlined by Funnell (Chapter 8), maintains that the principle of modularity relates to the knowledge requirements and proces- ses demanded by different tasks, not to the tasks themselves. When this approach is applied to brain damaged subjects, we would expect that damage could selectively affect certain types of knowledge or certain processes. Two tasks will dissociate only to the extent that they tap the same damaged or intact processing modules. It is also possible to outline a rather different interpretation of modular- ity, apparent amongst at least some computer vision researchers. For instance, Stevens (Chapter 2) considers the view that any two computa- tions may be thought of as being modular to the extent that they can be performed independently of one another. One implication of this view is that two tasks may be considered modular even though both tasks require
  • 29.
    *he same typeof process. Stevens cites the example of surface interpola- tion. Our ability to see a smooth surface at a different depth to its background when presented with a random-dot stereogram (Julesz, 1971) requires the interpolation of this surface between corresponding points in the left and right retinal images. Also, our ability to derive 3-D form information from motion requires the interpolation of a smooth surface between points and edges in the image where velocity profiles can be computed. It is possible that two-surface interpolation procedures are instantiated in two different modules: One concerned with deriving 3-D form information by stereopsis and one concerned with deriving 3-D form information from motion. The job of such modules would be to make available (independently), 3-D form descriptions from different types of stimulus information. Such descriptions would then be combined via a further, subsequent process. The implications of this second view of modularity for understanding neurological deficits in performance are quite different to the implications of the first. Here, we would expect brain damage to selectively affect abilities such as the computation of 3-D form from stereo information, or even a process such as surface interpolation in stereopsis. However, extensive damage to the “ stereopsis module” may nevertheless leave the “form from motion module” (including its surface interpolation component) intact. Present knowledge of the modules mediating visual object processing is probably insufficient to decide between these two views. Indeed, it may even be that the type of modularity governing early visual processing differs from that governing “ higher-order” processes such as recognition and identification. The implications of the differing views of modularity should, however, be borne in mind when considering the kinds of approaches taken when investigating selective visual processing disorders. 2.4. Interactions between Modules In discussing the principle of modularity, Marr (1982) notes that this principle “ does not forbid weak interactions between different modules” . The problem here from a behavioural point of view is in defining the extent and nature of any interactions. At one extreme we might presume that processing is organised in a series of discrete stages, with information being made available to each subsequent stage only following the resolution of processing at an earlier stage. Such discrete stages can be described as being strongly modular, in the sense that each processing stage is immune from other stages (e.g., from the influence of partial information from earlier stages or from information fed forward from later stages). At another extreme, one could imagine a system in which no processing stage is immune from the influence of any other. This can be said of “ interactive 8 HUMPHREYS & RIDDOCH
  • 30.
    1. INTRODUCTION 9 activation”models of information processing, in which partial information is transmitted between processing stages and in which later stages feed forward to constrain earlier stages (McClelland & Rumelhart, 1986). Such models are only weakly modular. For instance, brain damage could selec- tively isolate an earlier process from a later process, even though in normality the processes would interact (see Patterson & Morton, 1985, for a similar distinction concerning models of visual word pronunciation). Again, models differ in terms of whether they take a relatively strong or weak view on this topic. Marr’s (1982) view is characteristically strong; the connectionist models discussed by Humphreys and Quinlan and the cas- cade models of picture naming discussed by Riddoch and Humphreys (both in this volume) are relatively weak. Different accounts of brain damaged patients may follow from the contrasting assumptions about modularity. For instance, a strongly modular account of picture naming might hold that the structural characteristics of objects are processed fully before later information about, say, an object’s name is contacted. Accord- ing to this account, a patient who makes visual errors in naming pictures would be thought to have a problem in processes up to the stage of accessing structural knowledge about objects. An account assuming only weak modularity (e.g., the cascade model; Riddoch & Humphreys, Chap- ter 4) would not necessarily take the same position, since stages subsequent to that involving access to structural knowledge about objects could be affected by partially activated structural information. A patient could make visual errors due to problems in these later stages (see Riddoch & Humphreys, 1987, for a more detailed discussion). Additionally, the inferences we make about normal processes, using evidence from patients, are influenced by assumptions about modularity. In a strongly modular system, selective damage to one process would leave any functionally independent processes intact. In a weakly modular system this may not occur, since a problem in one process could constrain the operation of others. Such a conclusion is possible whether one’s account is pitched at an algorithm-level description or at a description at the level of the hardware. Sergent (Chapter 5) is concerned with the functions of the left and right cerebral hemispheres in object perception (a hardware-level issue). She argues that the capabilities of the left and the right hemisphere to process high and low spatial frequency information may differ. For example, the right hemisphere may preferentially operate on lower spatial frequencies than the left. Following unilateral right hemisphere damage, a patient may fail on a task not because the processes required for perfor- mance are functionally lateralised in the right hemisphere (e.g., the proces- ses required to match objects across viewpoints; cf. Warrington, 1982; see also Humphreys & Quinlan, Chapter 3), but because right hemisphere damage changes the quality of the information available to the left hemis-
  • 31.
    10 HUMPHREYS &RIDDOCH phere. In normality, task performance may be critically dependent on the co-activation of information in both hemispheres. In this case, we would be wise to be cautious when using single pieces of evidence from brain damaged patients to infer the localisation of one behavioural function either within a single processing module or neural structure. For instance, our conclusions about the nature of any processing module or its anatomi- cal locus would be bolstered by converging evidence that a deficit was functionally localised: by using tasks tapping other processing stages, by examining the temporal characteristics of processing in more detail, or by simulation. 2.5. Positive Attributes The problems involved in taking a cognitive neuropsychological approach to visual object processing thus concern issues about the kinds of descrip- tion that are most appropriate for explaining behaviour, about the relations between the different descriptions, about the kinds of processes which serve as the functional modules within the processing system, and about the relations between modules. Overall, it seems likely that visual object processing will best be understood by detailed examination of different kinds of description and of the contrasting questions which arise. As our knowledge about performance expressed at these different levels increases, we may hope for increasing convergence between such apparently contras- ting fields of study as computer vision, experimental psychology, and neurophysiology. The possibility of constraints between these different fields represents one of the most exciting areas of future development, and one of the strengths of the cognitive neuropsychological approach. One of the probable consequences of developments in the field is that we will produce more detailed models of the component processes involved in different tasks. Such models in turn produce more detailed predictions of the kinds of performance dysfunction which may follow brain damage. Disorders specific to visual object processing are typically classed within the syndrome of visual agnosia, of which two forms are distinguished: apperceptive agnosia, thought to pertain to the perceptual representation of the visual world, and associative agnosia, thought to pertain to the association of perceptual information with world knowledge (see Kertesz, Chapter 6). However, as our knowledge of visual object processing from other areas grows, it seems increasingly unlikely that a simple two-stage account of processing and its breakdown will suffice. The initial “ nosolog- ical” distinction between apperceptive and association stages (cf. Marin, Chapter 9) will fractionate, to provide a more precise account of the similarities and differences between patients. By bringing together work from different research fields, we hope to indicate the ways in which new
  • 32.
    1. INTRODUCTION 11 distinctionsabout visual object processing in general are developing (e.g. Stevens, Humphreys & Quinlan, Sergent, Riddoch & Humphreys, Funnell, this volume), and the ways in which such distinctions are being applied to understanding agnosia (e.g., Kertesz, Marin, Humphreys & Riddoch, this volume, Chapter 10). The utility of the multi-stage view which appears on the horizon (see Kertesz, Humphreys & Riddoch, this volume) can only be judged in time, in terms of how well it accounts for patients and of how useful its distinctions are for accounting for normal performance. In the meantime, it may serve as a fresh focus for work in the area. 3. THE BOOK The book is divided into two sections. In the first section, the chapters approach visual object processing from a consideration of which processes may need to be adopted either to develop an intelligent artificial visual system or to explain normal performance. Stevens (Chapter 2) discusses object processing from the perspective of computer vision. He outlines the kinds of assumptions adopted by workers in this field concerning functional modularity and the implementa- tion of the knowledge into procedural rules within the modules (such as “ rigidity” assumptions in deriving 3-D form information from motion). He goes on to consider some of the implications of computer vision for understanding both normal vision and its breakdown following brain damage. The following chapters by Humphreys and Quinlan (Chapter 3) and by Riddoch and Humphreys (Chapter 4) focus on particular operations in visual object processing: Object constancy and the processes involved in accessing stored structural, semantic and name information about objects. Humphreys and Quinlan’s chapter draws on work from a number of fields, such as computer vision, experimental cognitive psychology, animal vision, and neurologically impaired human vision. After considering the kinds of logical problems which constrain theories of object constancy, evidence concerning the mechanisms involved is reviewed. Humphreys and Quinlan propose that the problems of object constancy are not solved in any single way by the visual system, but that different solutions are adopted depend- ing on the kinds of information available and the uses to which that information must be put. They suggest that future research would prosper by de-emphasising general purpose solutions to constancy and by consider- ing in more detail stimulus and task constraints. Riddoch and Humphreys (Chapter 4) are concerned with the higher- order processes involved in accessing stored knowledge for picture naming. Various accounts of picture naming in normal observers are outlined, and a framework is developed which holds that picture naming operates in
  • 33.
    cascade. The cascademodel is then applied to account for different forms of picture naming disorder in patients. This chapter provides an example of a “weakly modular” account of visual object processing (see Section 2.4 above), and it attempts to explain both why such an account might be needed and its implications for understanding neurologically impaired patients. The final chapter in this section of the book, by Sergent (Chapter 5), focuses primarily on the anatomical localisation of processing mechanisms in the left and right cerebral hemispheres. However, in attempting to do this, Sergent draws on evidence from normal observers and animals in addition to brain damaged humans. It again provides an illuminating example of how we may look to evidence from seemingly different areas to further our understanding of vision. Necessarily, the chapter highlights the issue of-face perception, since considerably more detailed work on hemis- pheric differences in face perception has been conducted than on such differences in object perception (particularly in terms of developing the kinds of process models necessary to explain both abilities). Nevertheless, many of the lessons to be learned are the same, since faces are perhaps but one extreme example of objects which require accurate and integrated visual information for recognition to occur (see Diamond & Carey, 1986). The importance of viewing normal performance as the product of interac- tions between different processes is emphasised (see Section 2.4). The second section of the book deals with the types of visual object processing disorder which may occur in human subjects after brain dam- age. Kertesz (Chapter 6) outlines the historical background of work in this area. He describes which behavioural deficits are typically labelled as apperceptive or associative agnosia, or as other forms of visual object processing disorder such as optic aphasia; he considers the relations between these disorders and some associated deficits (such as colour and face perception, word recognition, etc.); and he considers the utility of a two-stage account of agnosia (see earlier). Kertesz’s review provides a useful introduction to agnosia from a clinical perspective. Campion (Chapter 7) focuses on the visual processing disorder typically labelled as apperceptive agnosia. His particular concern is with how psychological constructs (such as “perceptual representation” ) should be operationally defined to help understand the deficits in such patients, and on whether distinctions between “ sensory” and “ perceptual” processes are useful in this respect. He discusses how, in at least some instances, it might be useful to group patients according to the type of neurological damage sustained, rather than on the basis of higher-order constructs. Insights might then be gained by considering how the neurological damage con- strains information processing. This approach stresses how “ hardware- level” descriptions can further our understanding of the processing mechanisms mediating task performance. 12 HUMPHREYS & RIDDOCH
  • 34.
    1. INTRODUCTION 13 Funnell(Chapter 8) discusses more central deficits in visual object processing, particularly those concerned with the involvement of semantic knowledge. She outlines a novel approach which uses evidence from word matching, naming, and comprehension tasks with patients to address the issue of the nature of the semantic information mediating both these tasks and visual object processing. The relations between our stored knowledge of different properties of objects (such as sensory and functional know- ledge) are also considered, using data drawn from the ratings of normal subjects. Views of the nature of the semantic system are clearly important for understanding central deficits in visual object processing. Funnell argues that the semantic representations of objects differ in terms of the types and number of attributes which are specified, and that the loss of particular types of attribute may be important in understanding certain types of dysfunction. Central deficits in visual object processing are also of concern to Marin (Chapter 9). He deals with the physiological changes to the brain which occur in various forms of dementia, and with how such changes may precipitate deficits in object recognition and naming. Of particular interest is the fact that dementia is often consequent on quite different physiologi- cal changes to those observed in patients with brain lesions following head injury, stroke or tumour. The question is raised whether accounts of the processing deficits sustained also need to reflect these physiological differ- ences (see Section 2.2). Irrespective of whether one is sensitive to such hardware-level effects, cases of dementia can provide examples of remark- ably spared islets of ability amongst otherwise gross behavioural distur- bances; and it seems likely that work with such cases will be increasingly important for understanding the breakdown of what are, in normality, highly-interactive processes. In the final chapter we (Humphreys & Riddoch) attempt to examine in detail some of the implications for classifying visual agnosia in terms of a multi-stage theory of visual object processing. An expanded classification scheme is outlined to account for documented cases of agnosia, and to predict further ways in which object processing might break down. The aim is to provide a more flexible classification scheme which enjoys a better fit with theories of normal vision and which can be used predictively. Consid- eration is also given to the relations between different behavioural deficits and different types and sites of lesion. We hope that the differing fields drawn upon in this volume will attract readers from contrasting backgrounds and that each will find something of interest, not only from within their own field. We also hope that the approach adopted illustrates some of the ways in which we may, in future, cross the boundaries of different disciplines to gain a fuller understanding of human behaviour.
  • 35.
    14 HUMPHREYS &RIDDOCH REFERENCES Ballard, D. H., Hinton, G. E., & Seinowski, T. J. (1983). Parallel visual computation. Nature, 306, 21-26. Beck, J. (1967). Perceptual grouping produced by line figures. Perceptual & Psychophysics, 1, 371-394. Beck, J. (1982). Texture segregation. In J. Beck (Ed.), Organization and representation in perception. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Coltheart, M. (1984). Editorial in Cognitive Neuropsychology, 1, 1-8. Coltheart, M. (1985). Cognitive neuropsychology and the study of reading. In Μ. I. Posner and O. S. M. Marin (Eds.), Attention and performance, Vol. XI. Hillsdale, N.J. Lawrence Erlbaum Associates Inc. Cowey, A. (1985). Aspects of cortical organization related to selective attention and selective impairments of visual perception. In M.I. Posner and O.S.M. Marin (Eds.), Attention and performance, Vol. XI. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Desimone, R., Schein, S. J., Moran, J., & Ungerleider, L. G. (1985). Contour, color and shape analysis beyond the striate cortex. Vision Research, 25, 441-452. Diamond, R. & Carey, S. (1986). Why faces are and are not special: An effect of expertise. Journal of Experimental Psychology: General, 115, 107-117. Feldman, J. A. & Ballard, D. H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254. Hinton, G. E. & Anderson, J. A. (1981). Parallel models o f associative memory. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc. Hubei, D. H. & Wiesel, T. N. (1968). Receptive fields and functional architecture of monkey striate cortex. Journal of Physiology, 195, 215-243. Julesz, B. (1971). Foundations of cylopean perception. Chicago: University of Chicago Press. Julesz, B. (1980). Spatial nonlinearities in the instantaneous perception of textures with identical power spectra. Philosophical Transactions of the Royal Society, London, B290, 83-94. Julesz, B. (1981). Textons, the elements of texture perception and their interaction. Nature, 290, 91-97. Julesz, B. (1986). Texton gradients: The texton theory revisited. Biological Cybernetics, 54, 245-251. Land, E. H. (1977). The retinex theory of color vision. Scientific American, 237, 108-128. Marr, D. (1976). Early processing of visual information. Philosophical Transactions o f the Royal Society, London, B275, 483-524. Marr,D. (1979). Representing and computing visual information. In P. H. Winston and R.H. Brown (Eds.), Artificial intelligence: An MIT perspective. London: MIT Press. Marr, D. (1980). Visual information processing: The structure and creation of visual representations. Philosophical Transactions o f the Royal Society, London, B290, 199-218. Marr, D. (1982). Vision. San Francisco: W. H. Freeman. Marr, D. & Nishihara, H. K. (1978). Representation and recognition of the spatial organiza- tion of three-dimensional shapes. Proceedings o f the Royal Society, London, B200, 269-294. McLelland, J. L. & Rumelhart, D. E. (1986). Parallel distributed processing: Explorations in the microstructure ofcognition, Vol. 2: Psychological and biological models. London: MIT Press. Mehler, J., Morton, J., & Jusczyk, P. W. (1984). On reducing language to biology. Cognitive Neuropsychology, 1, 83-116.
  • 36.
    Patterson, Κ. Ε.& Morton, J. (1985). From orthography to phonology: An attempt at an old interpretation. In Κ. E. Patterson, J. C. Marshall, and M. Coltheart (Eds.), Surface dyslexia: Neuropsychological and cognitive studies of phonological reading. London: Lawrence Erlbaum Associates Ltd. Quinlan, P. T. & Humphreys, G. W. (1987). Visual search for targets defined by combina- tions of color, shape, and size: An examination of the task constraints on feature and conjunction searches. Perception & Pyschophysics, 41, 455-472. Riddoch, M. J. & Humphreys, G. W. (1987). Visual object processing in optic aphasia: A case of semantic access agnosia. Cognitive Neuropsychology, 4, 131-186. Sutherland, N. S. (1979). The representationof three-dimensional objects. Nature, 278, 395-398. · Treisman, A. & Gelade, G. A. (1980). A feature integration theory of attention. Cognitive Psychology, 12, 97-136. Treisman, A. & Paterson, R. (1984). Emergent features, attention and object perception. Journal of Experimental Psychology: Human Perception and Performance, 10, 12-31. Treisman, A. & Souther, J. (1985) Search asymmetry: A diagnostic for preattentive proces- sing of separable features. Journal o f Experimental Psychology: General, 114, 285-310. Warrington, E. K. (1982). Neuropsychological studies of object recognition. Philosophical Transactions o f the Royal Society, London, 298, 15-33. Zeki, S. M. (1980). The representation of colours in the cerebral cortex. Nature, 284, 412-418. Zeki, S. M. (1983). Color coding in the cerebral cortex: The responses of wavelength- selective and color-coded cells in monkey visual cortex to changes in wavelength composi- tion. Neuroscience, 9, 767-781. 1. INTRODUCTION 15
  • 38.
    Visual Object Perception froma Computational Perspective Kent A. Stevens Department of Computer and Information Science, University of Oregon, Eugene, Oregon 97403, U.S.A. 1. INTRODUCTION David Marr introduced a computational approach towards the study of biological information processing, which stresses that a complex system such as the visual system is feasibly understood only across several levels of abstraction (Marr, 1982; Marr & Poggio, 1977). The suggestion is that no single level of description, be it behavioural, psychophysical, computa- tional, or neurophysiological, can adequately describe both function and implementation with sufficient detail and scope to allow the necessary breadth of understanding. The necessity for multiple levels of description is accepted in computer science without question. When designing or analysing a complex informa- tion processing system one attends either to questions of what the require- ments, goals, and tasks of the system are, or to questions of how the system achieves these goals, such as what the algorithms are and how the data are organised.1 One shifts between these two levels of description often *An algorithm, as defined in computer science, is the formal specification of a particular computation. Informally, it is a sequence of step-by-step instructions or commands which, if interpreted by some processor, would result in the achievement of some computation. An algorithm is expressed by means of a language having an agreed-upon set of constructs for referring to data structures, operations on data structures, and constructs for controlling the execution of these operations. In principle, one can define parallel as well as sequential algorithms, for a variety of types of processing mechanism ranging from analogue to digital 17 2
  • 39.
    without regard fora third, more detailed level, that of the actual hardware on which the algorithms run and the data reside. These three levels of description are rather distinct in computer science, and each has its appropriate use and limitations. It is simply not feasible to describe a complex computation in terms of events at the level of the hardware. The technical language and formal principles that are relevant at the level of the digital hardware are neither appropriate nor relevant at the level of algorithm and program, and vice versa. Descriptions of computations are best made via a technical language and a set of formal principles concerned with algorithms, data structures, and other relevant abstractions. It would be theoretically possible to describe the precise pattern of hardware events arising within a computer during, say, the computation of a least squares fit algorithm. The determinism of the digital computer allows, with considerable tedium, the complete specification of its detailed behaviour down to the most elemental state transitions within its electronic components. However, consider working upwards, inferring what the processor is doing, what function it is performing, purely on the basis of the observed hardware event. It is not merely a matter of the overwhelming complexity which makes this not feasible, it is the inappropriateness of the technical language for other than the description of local events. Different languages have been developed for describing computations at different levels of specificity. Of course, computer science is a synthetic science involved with the development and study of computational machinery as it can be humanly conceived and understood. One might argue that limitations in our ability to conceive of and understand complex systems fairly forces one to attend exclusively to either the higher (algorithmic) level or the lower (implementation) level. Be that as it may, one might also argue that computer science has found a fundamental principle of computation (a law perhaps as fundamental as entropy or the conservation laws in physics), namely, that one can describe a computation independently of how it is implemented, or equivalently, that there are arbitrarily many equivalent implementations of a given computation. This 18 STEVENS (discrete state). When one forwards a particular algorithm one suggests a particular recipe for how the computation proceeds and in what form the data to be manipulated are represented. There are infinitely many equivalent algorithms for achieving any given computation (con- sider how many ways one could rewrite a computer program which would not effect its external behaviour). The choice of algorithm depends, in part, on issues of how the given primitive representations and functions are implemented—the machinery at hand. It also depends on design goals such as fault tolerance or “graceful degradation” of behaviour as input is impoverished, time and space efficiency, the immediacy with which the output of the computation is provided to subsequent computations, and so forth (Marr, 1982, p. 106). Thus, when discussing a computation at the level of algorithm one uses rather specific notions of data structure and the operations that transform input into output.
  • 40.
    principle, established byTuring and others earlier in this century, has provided the underpinnings for the formal description of computations. Marr suggests that there is no fundamental reason why this principle does not apply equally to biological information processing, and to vision in particular. That is, vision is achieved by processes that extract and manipulate visual information. These computations, while certainly not implemented in the same manner as those performed in silicon technology, are computations nonetheless. Furthermore, being computations, they are amenable to a formal description. One question, then, is what advantage is gained by describing human vision formally as computations? Another question is whether that formal description is really able to be divorced from the biology, or whether it is a nicety of computer science applied to computers, but not of computer science applied to organisms. I think that the answers to both are to be found in the computational work that has been performed in the last decade or so. The purpose of this chapter is to indicate the utility of this approach and to demonstrate where the extensions of these results will take us. 2. MODULARITY IN VISION If vision is to be characterised in terms of information processing problems, how do we determine what any specific problem is in order to study it? Consider colour perception, stereopsis (the creation of depth information from the disparities between the images in the left and right eyes) and the extraction of three-dimensional (3-D) information from motion, all of which have clear computational goals. In principle, one could imagine an independent solution to each, and behaviourally, there appears to be substantial independence of function. The dogmatic view of functional modularity would be that if two computations are theoretically orthogonal (i.e. logically able to be per- formed independently), they should be treated as distinct even though they may share many common underlying mechanisms. For example, the per- ception of colour seems orthogonal to the problem of determining the direction of motion in the retinal image. Consider then the colour- contingent motion and after-effect. One defers the question of how or why such couplings occur until later, when questions of visual mechanism are examined. A reasonable sceptic might regard this approach as dangerously simplis- tic. How can we trust our intuitions as to what computational problems are distinct and which are inherently interrelated, and how can we tell which visual phenomena reflect the visual algorithms and which are merely due to “ implementation details” ? More generally, how can we tell whether our 2. VISUAL OBJECT PERCEPTION 19
  • 41.
    decomposition of visioninto modules reflects that which nature chose? The still more sceptical might further suggest that it is misguided to imagine any modular decomposition of vision, that vision can be understood only by embracing its complexity and interactions as a whole. I would remind them that we have already developed computer systems the complexity of which defy understanding except in terms of levels of abstraction (ignoring the hardware when thinking of the functionality, and vice versa) and only very locally (considering only a very narrow aspect of the system under the simplifying assumption that there are no important dependencies on global states of the system that are not already reflected by local parameters). Of course, we return to the fact that computer systems are artefacts that are intentionally made with a high degree of strict modularity—we cannot make complicated systems otherwise. Marr (1982, p. 102), however, suggests that nature has also adopted the principle o f modular design, primarily because if the components of a system are not made as independent of one another as the individual tasks allow, a small change to one component will have undesired consequences elsewhere. A system that is not modujar would be difficult to improve or extend; evolution presumably favours a functionally modular design. Functional modularity does not mean strict segregation of function at the neural level, however. Actual segregation or isolation of processes, in fact, has its disadvantages, primarily poor economy of design and the problem of communicating functionally useful information among processes. If two problems are interrelated (such that the solution to one partly constrains the solution to the other), it would be disadvantageous to isolate them. Moreover, just as evolution favours functional modularity, it probably is limited in how literally the modularity can be extended. That is, it would seem improbable that evolution would extend the capabilities of a visual system by implementing the new perceptual ability with neural structures that are strictly isolated from the neural mechanisms that preceded it. More likely would be either the duplication or subdivision of existing cortical structures in order to perform some new, ecologically advantage- ous, computation on existing internal information structures (see, for example, Livingstone and Hubei, 1984, where they suggest that new functional architecture to support colour vision was introduced into the centre of the h]/percolumns found in striate cortex). As a consequence, the new functional architecture would not only be intimately tied with the earlier architecture, it would probably incorporate facilitatory interactions between related processes in an ad hoc and case-by-case manner. I would doubt that evolution adheres to strict design standards for adding new functional capabilities; it is probably governed primarily by the particular problems at hand. For example, consider the problem of combining the 3-D information 20 STEVENS
  • 42.
    2. VISUAL OBJECTPERCEPTION 21 delivered simultaneously by stereopsis and motion. Both motion and stereopsis have limitations in the quality of the 3-D information they can deliver individually. In fact, they are complementary so that the results of each process can be combined to provide information that is otherwise ambiguous (Richards, 1985; Waxman & Ullman, 1985). It is not known, however, whether the combining of motion and stereopsis is achieved at a later stage of depth processing, with no “ cross-talk” between the two processes that deliver depth information, or concurrently by more intimate interaction between the two. Since there are, in addition to motion and stereopsis, several additional sources of 3-D information, it would seem easier to imagine a computational architecture for combining 3-D informa- tion that minimises interactions between or among the contributing “ mod- ules” . As to what global architecture is incorporated in human vision is still largely a matter of speculation. Most of the insight has been gained at the level of the brain’s ‘‘microarchitecture’’, as it were, the processing achieved by individual neural units selective to, for example, motion or stereo disparity. 3. SURVEY OF COMPUTATIONAL PROBLEMS In the following we discuss several problems central to spatial perception: stereopsis, the 3-D interpretation of motion, the interpolation or “ filling in” of smooth surfaces, the description of surface shape, the 3-D interpre- tation of contours and texture gradients, and very briefly sketch the processing that might follow the perception of visible surfaces. In examin- ing a variety of information processing problems such as these one can discern a recurring theme in this type of work— one of discovering the physical (or geometrical) constraints that allow the solution to a given visual problem, postulating the method by which the constraints are incorporated into some computation, making explicit various aspects of how the data are represented and processed, and eventually accounting not only for psychophysical results but tying these suggestions to particular neuorophysiological mechanisms. It is a relatively recent idea to attempt to bridge between computationally motivated theories and clinical findings. After the following survey we will close on that note. 3.1. The Matching Problem in Stereopsis A particularly clear problem is presented by the matching problem in stereopsis, namely the determination of the correspondence between left and right images in order that the angular disparity between corresponding elements may be registered. Each image point in the one eye might correspond to any of a range of image points in the other eye, depending on where the point projects when viewed from the perspective of the other
  • 43.
    eye. This hasalso been termed the false-targets problem. For instance, given an image with 4 points in the left eye and 4 points in the right eye, there are 16 possible matchings, only 4 of which are correct, the remaining 12 being false targets. How is the correct matching established over a rich, detailed image where false targets abound? Various proposals could be made as to the image features that are brought into correspondence— intensity points, segments of edges or lines, and more complicated features involving colour, contrast magnitude, and so forth. For example, one might propose local correlation of the intensity values to estimate the stereo disparity in each vicinity. However, is the computational problem of stereopsis being addressed appropriately in such proposals? Does intensity-value correlation solve the underlying computa- tional problem? Empirically, correlation of individual pixels (intensity samples) has generally proven inadequate. The problem lies in the approp- riateness of the tokens that one seeks to match across left and right images. Simply put, there is no reason to expect a pixel-to-pixel correspondence across images. In contrast, Marr and Poggio (1977) recast the computa- tional problem of stereo matching more as one of matching corresponding physical surface features, or at least, of matching elements in the image that have a more reliable tie to surface features than mere pixels. One consequence of this view was to rethink the basis for the matching computation and to derive three constraints on the matching problem. One isthecontinuity constraint. Because the visual world isdominated by smooth, op ^ue surfaces, stereo disparity varies smoothly almost everywhere, be ause the distance to the visible surfaces varies continuously except at surface boundaries. Note that it is precisely at surface boundaries where the correlation approach would fail. Additionally, Marr and Poggio recog- nised two additional constraints on the matching problem—compatibility, which means that the descriptive elements in each eye must be compatible with being adjacent views of the same physical features (hence the contrast signs must match, and edges must match edges, etc.), and uniqueness, which means that the correspondence must be one-to-one almost every- where (the exceptions occurring along occlusion boundaries). Marr (1982, p. 114) reflects on these constraints and suggests the following fundamental assumption o f stereopsis. If a correspondence is established between physically meaningful primitives extracted from left and right images of a scene that contains a sufficient amount of detail, and if the correspondence satisfys the three matching constraints, then that correspondence is physically correct (and it follows that the correspon- dence is unique). Marr suggests that these three constraints are sufficient to solve the stereo matching problem. Identifying, justifying, and demonstrat- ing the sufficiency of such a set constraints is the central activity of deriving a computational theory. We will see other examples momentarily. Note 22 STEVENS
  • 44.
    2. VISUAL OBJECTPERCEPTION 23 that while the constraints seem reasonable and plausible it is not clear that much headway has been made on the problem at hand, namely explaining how the visual system solves the stereo matching problem. After all, one must get back to worrying about matching corresponding items in the two images, and terms such as “ physically meaningful” do not immediately suggest particular matching primitives. In fact, it is when one attempts to become more specific that one inevitably goes “ down” a level of abstrac- tion, down to the level of algorithm. Marr and Poggio have examined two quite different algorithms for stereopsis (Marr, 1982), and Mayhew and Frisby (1981) independently proposed similar, highly plausible algorithms vis-à-vis human stereopsis. The insight that produced these computa- tional theories of human stereopsis came from thinking about the underly- ing physical constraints that might, in principle, admit a solution. It should be pointed out that a variety of computational theories for matching have been proposed, each based on a different combination of constraints. Mayhew (1982) describes five constraints that are more or less independent and which can lead to different matching schemes having different performances. The constraints are: (1) the opacity constraint; (2) the epipolar constraint, that a plane passing through the optical centres of the eyes projects to a straight line in each eye, leading to strategies for searching along particular loci for matches; (3) figurai continuity, that edges of surfaces and surface markings are spatially continuous; (4) an object in a given point in space will project to luminance changes in the two eyes with similar descriptions; and (5) the ordering constraint, wherein the stereo projection almost always preserves the ordering of luminance changes along corresponding paths. Mayhew (1982) then shows how different combinations of these constraints have been incorporated in different theories and discusses the relative power of these constraints (in enforcing particular matches or excluding certain matches locally). A computer implementation can be made of a particular algorithm which embodies a given set of these matching constraints, and its behaviour examined when given digitised images of stereo pairs. It is possible to digitise the image with sufficiently fine resolution and broad dynamic range to match the granularity and range of the retinal receptors, then filter the digital image by local lateral-inhibition operators that model retinal gang- lion X-cells. The receptive field of these operators is modelled as a difference of Gaussian distributions (see Marr, 1982; Marr & Hildreth, 1980; Wilson & Giese, 1977). With an implemented model of the early intensity processing and for the specific image features of tokens that are matched between left and right images, one can implement and test a given stereo matching algorithm. The performance of the algorithm can be compared with human vision across a wide range of visual stimuli, from random dot stereograms to natural images. The “ psychophysics” of a
  • 45.
    24 STEVENS computer implementationof a stereo matching theory is rather difficult to interpret. If the implementation utterly fails to achieve the same stereo correspondence as do humans in viewing a particular stereo pair one can surmise that the two methods differ in some combination of the matching constraints or the matching tokens, (for example, Mayhew and Frisby, 1981, devised an ingenious stereo pair that tested a detailed prediction regarding the stereo matching tokens incorporated in human vision). It is more difficult to interpret the behaviour of an implemented theory when it is reasonably close to human behaviour but deviates consistently and significantly. One interpretation of the performance disparity is that the human visual system embodies a close, but not exact, engineering approx- imation to the process laid out in the given theory. That is, the theory is correct but does not capture all the implementation details. The alternative explanation is that one does not yet have the right theory, of course. There have been several stereopsis implementations by different investigators, each based on similar but slightly different combinations of computational constraints. The evolution of theories has led to increasingly close matches of predictions of the known psychophysics of stereopsis, but a conclusive demonstration of the validity of a given theory solely on the basis of similarity of behaviour has remained illusive. The basic problem of stereopsis, recall, is determining the correct match between the two images, so that stereo disparity may be computed. There are subsequent computational problems associated with the interpretation of the disparity information, foremost being the computation of distance from disparity (see Mayhew & Longuet-Higgins, 1982). Another problem, which has been considered primarily in the context of stereopsis, but which likely has wider scope, concerns the interpolation or reconstruction of a smooth surface between the points where stereopsis is able to provide depth information. In a sparse random dot stereogram, for instance, once the stereogram is fused, the points appear in space to lie on an apparently continuous surface that exists between the dots, as if the dots were embedded in atransparent sheet. Because interpolation phenomena can be seen also in surfaces depicted by motion sequences, we will return to it as a computational problem in its own right, after considering problems involved in the perception of space from motion. This interweaving of computational problems is commonplace and probably inherent to the nature of vision, and while it makes it difficult to develop clearly demar- cated theories, it is possibly unavoidable. Let us turn to the next topic of our survey, the 3-D interpretation of motion across the retinal image. 3.2. Motion Interpretation Problems The stereo matching problem was one of the first computational problems in vision to be addressed in detail. Another computational problem which has recently received at least as much attention concerns the 3-D interpre-
  • 46.
    2. VISUAL OBJECTPERCEPTION 25 tation of motion. Gibson (1950) wrote of the spatial information provided by motion parallax, and showed that in the instantaneous velocity field cast on the retina there is information about the layout of surfaces relative to the observer. That information has generally been regarded as either distance or local surface orientation. What earlier was regarded as merely motion parallax has been decomposed more recently into many computa- tional problems, depending on whether one is concerned with continuous vs. discrete motion, orthographic2vs. perspective projection, and so forth. (For the case of continuous motion in perspective projection see Koen- derink & van Doorn, 1976 and Longuet-Higgins & Prazdny, 1980; for continuous motion in orthographic projection see Hoffman, 1982; and for case of discrete motion in orthographic and perspective projection see Ullman, 1979.) Koenderink and van Doom’s (1976) work demonstrated how the optical velocity field3 contains information about the spatial distribution of sur- faces about an observer as one moves. The local velocity field in the image was shown to be decomposable into quantities that are related to depth, surface orientation, and velocity. Ullman (1979) examined the minimal information from which 3-D structures could be derived, and, for instance, was able to show that three distinct orthographic views of four non- coplanar points in a rigid configuration are sufficient for the recovery of their spatial arrangement, up to a reflection. That is, rigidity of motion results in a unique interpretation under those viewing circumstances. Again, assuming rigidity, Tsai and Huang (1984) have shown that, with certain exceptions, two perspective views of seven points are also suffi- cient. Various other results have been shown for restricted sorts of motion, such as motion about a planar or fixed axis (Bobick, 1983; Hoffman & Flinchbaugh, 1982), which have relevance to the interpretation of articu- lated systems such as animals in locomotion. Rigidity provides the strongest mathematical constraint on the spatial interpretation of motion. In fact, much of the computational work in the last decade has concentrated on the fundamentals of interpreting motion in the abstract, as essentially a mathematical issue. It assumes that motion information is available in some usable form in an image, such as the instantaneous velocity field (retinal angular velocity provided for all visual directions). Whereas Gibson (1950) discussed the wealth of spaital informa- tion provided by the velocity field, the information in the idealised optic Orthographic, or parallel projection is approximated when the variation in distance across a region of a 3-D scene is small relative to the overall distance. 3 The 2-D array of velocity vectors across the retinal image induced by movement relative to the visual surrounds. The velocity field is presumably computed by motion detectors that deliver information about speed and direction of movement across the retinal image. The velocity field is then expected to be a rich source of information for subsequent 3-D analysis.
  • 47.
    array that isincident on the retina is quite difficult to extract or “ register” . It appears that computing the velocity field poses substantial problems in its own right, problems that certainly were not foreseen prior to recent computational experiments. For instance, the so-called “ aperture prob- lem” states that only the velocity component normal to an edge or line segment is locally computable, the component tangent to the edge or line cannot be determined by a motion detection operator having a limited aperture or receptive field (Marr & Ullman, 1981). In actuality perhaps only a rather sparse and perhaps crude approximation to the instantaneous velocity field is neurally determined, which would then pose severe prob- lems on any motion processes that use this velocity information as input. The perceptual processes that seek to make 3-D interpretations must be able to cope with errorfu1 and incomplete velocity information (see, for example, Ullman & Hildreth, 1983). While progress has been made regarding the measurement and interpre- tation of motion, the result of the 3-D interpretation, whether represented in terms of depth, surface orientation, or some other form, is still quite obscure. Part of the problem is that we have, as yet, little insight into how spatial information, however extracted, is organised within the human or animal visual system. Some of the issues that must be clarified are: precisely how distance information is encoded, whether distance is made explicit for all visible points across the seen surfaces up to some resolution, how surfaces are described over different scales (for instance, some natural surfaces are smooth on a coarse scale but quite irregular on fine scale, others are smooth in detail but jagged overall). These issues arise when attempting to actually account for the interpretation of motion in natural images, since the irregularities of the seen surfaces result in irregularities in the velocity field, just as smoothness of the surfaces results in smoothness in the velocity field. But the problems of surface description are not restricted to motion interpretation, and that leads us to the next point. 3.3. Surface Interpolation Problems It was mentioned earlier that stereopsis does not provide stereo disparity information at all points in the binocular image, but only at discrete places such as edges, lines, and points that have been brought into stereo correspondence. The intervening, featureless, regions can only be assumed to lie at similar depths if the surface is assumed to be smooth. To make a particular reconstruction of a smooth surface consistent with the points where the disparity is known is not a problem unique to stereopsis, however. Just as stereopsis brings only certain features into stereo corres- pondence, motion provides spatial information only at places for which velocity information can be attributed, such as edge and line segments, 26 STEVENS
  • 48.
    2. VISUAL OBJECTPERCEPTION 27 blobs, and points. Nonetheless, one can derive from a discrete pattern of moving dots the impression of seeing a coherent sheet of dots in 3-D, as if one attributes motion to the intervening regions between such features. To do so requires an assumption of viewing a smooth surface when the dots behave in a coherent manner. The same holds for surfaces suggested by contours, such as shown in Fig. 2.1. The smooth surface is depicted by discrete, separated contours, between which no surface information is explicitly available. The impression of an overall surface under the con- tours may be attributed to a surface interpolation process. Some computational attention has been paid to the problem of surface interpolation, the process of reconstructing a smooth surface between the discrete places where the surface is known (see, e.g., Grimson, 1981). Two fundamental research concerns are: (1) whether or not the human visual system explicitly interpolates an entire smooth surface in parallel across the image, and if interpolation exists in human vision: (2) to determine the interpolation strategy or strategies. Although it is not certain that explicit interpolation does take place in human vision, it has been assumed so, and work thus far has centred largely on formalising mathematical schemes for surface reconstruction analogous to the engineering practice of determin- ing smoothly faired surfaces. For instance, one can regard the interpolation as a smooth membrane that is stretched over the discrete places where the surface shape is known, then compute that particular surface shape for the membrane that minimised its elastic potential energy. Rather than use a membrane, an alternative approach is to model the surface as a thin plate, which is resistant to flexure and torsion, and which smoothly bends without fractures or creases to conform with the boundary constraints. Clearly one of the issues that must be dealt with theoretically is how a surface can be described simultaneously as being smooth overall but jagged or irregular in detail, or vice versa. The answer to this might result in different approaches to surface interpolation. One central question, it seems, is to what extent the visual system goes to the effort to explicitly compute multiple surface interpolations at various scales, the alternative being the computation of a surface discription, on demand as it were, for a particular scale of description, and only locally to a given patch of surface. A complementary problem to the interpolation problem is the discon- tinuity detection problem. That is, given a collection of discrete spatial points lying on different, spatially separated surfaces, how does one decide where to fit smooth surfaces and where to postulate surface boundaries? What constitutes a discontinuity, such as a crease or step edge between two smooth surfaces clearly depends on the scale one has in mind (e.g., a crumpled, then flattened sheet of aluminum foil presents both smoothness and jaggedness depending on the scale of reference). But consider the presumably simpler problem of deciding, for a given scale, whether or not
  • 49.
    28 STEVENS the surfaceis creased or smooth at a given point. One approach, for surfaces modelled as thin plates, is to locally measure the “ strain” on the surface and explicitly create discontinuities where the strain becomes excessive. 3.4. Describing Visible Surfaces and Objects Interpolation and the determination of surface creases and boundaries, as well as several other issues such as the description of local surface topogra- phy, are issues that span a broad range of surface perception processes. Parsimony would suggest that the solution to these problems are provided by central mechanisms of surface discription that are common to stereop- sis, motion, shading, texture, and so forth. Marr (1982) collectively termed the representation and associated processes that manipulate internal descriptions of visible surfaces the 2J-D sketch. The name suggesting that it is a sketch or representation of the salient spatial properties of the seen surfaces, not a full 3-D representation of the objects including the enclosed volumes and portions that are not visible from the particular viewpoint. The 2í-D sketch would achieve the integration of 2-D information from different sources such as stereopsis and motion, resulting in an at least locally consistent description of the surfaces that are visible at a given instant. It would also subserve the description of seen objects for purposes of object recognition, manipulation, and so forth. The quantitative description of object and surface layout had been central to computer vision efforts (see, Horn, 1977). Questions of 3-D representation arise in a very concrete manner when one seeks to describe viewed objects within a computer so that they might be automatically recognised, or grasped by a manipulator, or navigated around. However, unlike the proposed 2J-D sketch, in computer vision the representation question has concerned description of a single type of 3-D information derived from a single source (typically either shading or stereopsis) and questions of consistency with information generated from other processes, such as motion, have not been addressed. In contrast, from a psychological perspective there has been concern for the integration of visual cues, notably Attneave’s (1972, 1982) discussions of viewer-centred4 local descriptions of surface orientation and depth, and the intimate relationship between the representation of 3-D information and its extraction from 4A viewer-centred description is relative to the instantaneous perspective the observer has of the surrounding surfaces. Local surface shape and disposition is described relative to the line of sight to each surface patch, so that if the observer moves relative to the surface all values within the viewer-centred description might be expected to change. While volatile, such a description would be the basis for eventually deriving more stable object-centred descriptions (see discussion of axis-based schemes, p. 30).
  • 50.
    2. VISUAL OBJECTPERCEPTION 29 images. However, the 2|-D sketch proposal, and perhaps more impor- tantly its theoretical motivation, constitutes the broadest statement of the purpose of early vision, namely the processing of image information in order to develop a description of the visible surfaces (Marr, 1982). Vision achieves descriptions of the visual world, and towards that goal the 2|-D sketch would be one of the important representations within which these descriptions would lie. When phrased in such terms, one also sees clear parallels to Gibson’s (1950) theory that the visual system registers spatial information by means of “ higher order” variables in the optic array (the interested reader is referred to Ullman, 1980, for an excellent discussion of Gibson’s direct perception viewed from this perspective). The major contribution of proposing that early vision culminates in a description of the visible surfaces is probably that it causes one to question precisely what consti- tutes that description. There are many ways to quantify spatial information that are formally equivalent, but presumably only a few such ways are actually used within the visual system to encode such information. At some point in the course of investigation of surface and object perception these issues will have to be pinned down. Making precise and specific the nature and structure of the internal representations of surfaces and objects is extraordinarily difficult. The problem is difficult for many reasons, some of which are sketched in the following. First, one faces the thorny issue that there are indeed many equivalent ways of achieving an equivalent description, given that one cannot rule out specialised processes that access and manipulate the information contained in the representation. To illustrate, suppose the visual system has a 2-D buffer containing perceived depth as a function of visual direction (i.e. a depth map), and associated with it, the ability to compute the gradient of depth in a given locality. Since the magnitude and direction of the gradient of depth is straightforwardly related to the local surface orientation (see Stevens, 1983b), by having the ability to compute the gradient at any point one has, in effect, a local surface orientation map for all smooth surface points. Since a depth map facilitates the perception of local surface orientation, how then does one determine whether or not local surface orientation is made explicit within some 2-D buffer or simply computed locally “ on demand” ? The visual system is likely to be quite robust in its ability to derive different types of spatial inference as required by the visual task at hand. It is still an open question as to what primitive information is stored in a visual buffer and what information is subse- quently inferred or computed on that basis. Depth and local surface orientation constitute but two means for repres- enting surface shape. Other types of 3-D shape primitive can be proposed, such as measures of the local surface curvature (predicates indicating
  • 51.
    qualitative flatness orperhaps measures of the sign and magnitude of the Gaussian curvature and the principal directions). More extended topo- graphic features such as ridges and troughs, peaks and dips, etc., might be involved in the description of surface shape in human vision. A ridge or trough or an extremum in surface curvature is clearly visually salient, and it is reasonable to suspect that such features of surface shape are explicitly represented. From a computational point of view it is important to ask how a surface shape should be described, keeping in mind that the representa- tion of surface shape must subserve certain processing tasks that follow, such as object recognition, manipulation, and navigation. In the case of object recognition one of the important criteria of a representation (Marr & Nishihara, 1978) is how stable the description is over changes in viewpoint. Clearly the visual system should derive the same conclusion as to an object’s identity over as many different view- points as possible, with recognition difficulties arising for only particular vantage points where critical features are obscured, such as major axes of elongation or symmetry that are severely foreshortened (see Marr & Nishihara, 1978). Marr and Nishihara (1978) describe a hierarchical representation of objects that is particularly suited for articulated objects such as animals that can be regarded as the composition of parts. In this scheme each part of an object is described in terms of how the cross-section of the part varies as a function of position along its axis (an approach introduced by Binford, 1971). The connectivity among the component parts of an object is then described in terms of how the axes join, the relative sizes of the parts, etc. Marr and Nishihara (1978) observe that such descriptions capture the essence of a broad range of animal shapes, and demonstrate this fact with effective renditions of different animals by pipecleaners. That approach is most effective for describing objects made of connected parts that can themselves be described by axes. More recently, Hoffman (1983) has considered how to decompose a complex surface shape into component parts. The problem can be solved in part by considering the silhouette outline, and identifying the points along the silhouette where the parts intersect (see also Marr, 1977, concerning the interpretation of silhouettes, and the sketch of that work in Section 3.5.). Hoffman (1983; Hoffman & Richards, 1984) also extends these 2-D rules to 3-D, showing that surfaces can often be decomposed along extrema of negative principal curvature into physically meaningful parts. Note that in this scheme an object is described in terms of the intrinsic geometry of its surfaces and not in terms of volumetric primitives such as axes and cross-sections. Both schemes might well be incorporated in human vision, as these researchers demonstrate. Furthermore, the work pf Shepard and Metzler (1971) regarding the mental rotation of 3-D shapes strongly 30 STEVENS
  • 52.
    2. VISUAL OBJECTPERCEPTION 31 suggests that the visual system is able to manipulate in an effectively continuous manner an internal representation of a seen object which captures salient part-whole relations (see Shepard, 1981 for discussion). The connection between this ability and objection perception is not at all clear, unfortunately. 3.5. Interpreting Contours in 3-D The visual system is quite adept at imposing 3-D interpretations on line drawings, the line-drawn Necker cube being a familiar example. Although there are infinitely many 3-D configurations thatmight have projected to any given 2-D image, we tend to see a specific configuration in 3-D (and its reversal)— clearly, we impose strong constraints on the 3-D interpretation. It is well established that in viewing such figures we prefer that interpreta- tion which favours regularities, symmetries, parallelisms, coplanarities, etc. Consequently, we tend to see parallelograms and trapezoids as slanted rectangles or squares, trihedral vertices as right trihedral vertices, and so forth (see, e.g., Perkins, 1982; Shepard, 1981; Stevens, 1983a). These tendencies might be indirect consequences of more general Praegnanz principles or in some cases they might directly reflect specific perceptual assumptions, (e.g., of perpendicularity). To amplify this point, the Praeg- nanz principle (Koffka, 1935) says that we tend to choose the simplest interpretation, particularly in cases of minimal stimuli, hence given a trapezoid in 2-D we prefer the 3-D interpretation which, in addition to preserving parallelism and symmetry, results in equal length lines and equal angles at the vertices (which implicitly results in 90° angles in 3-D). While there is considerable evidence for Praegnanz in our visual interpre- tations, it does not exclude the possibility that we also drive our interpretations by strongly constraining assumptions that more directly entail perpendicularity and other geometric properties (discussed later). As to whether such geometric assumptions can be rephrased in terms of Praegnanz is probably a pursuit of secondary importance. In contrast, of considerable importance from a computational point of view is determin- ing precisely what 3-D interpretation problems exist, what each entails, and what constraint is employed to solve each. It is another matter to study how these various constraints are integrated, that is, by what methods do simultaneous and potentially conflicting constraints force one interpreta- tion over another. Questions of how these constraints might be implemented within some perceptual process can be studied distinctly from the constraints themselves. It is there that issues of Praegnanz re-emerge (see, for example, the discussion on “hill-climbing” , i.e. the method of gradient ascent, discussed by Attneave, 1982 and various proposed
  • 53.
    32 STEVENS computing architecturessuch as connectionist models, e.g., Ballard, Hin- ton, & Sejnowski, 1983; see also Humphreys & Quinlan, this volume, Chapter 3). In the following, we take the view that we can address the constraints that allow the interpretation of 3-D shape independently of how the constraints are implemented. The computer vision community examined the con- straint question in the limited domain of blocks (Waltz, 1975), showing how local configurations of vertices, while individually ambiguous, are mutually constraining so that a global 3-D interpretation can usually be derived, given assumptions of general position (ruling out accidental alignments) and a priori assumptions about the restricted domain (right trihedral vertices, planar facets, etc.). Note that the computer vision work is clearly related to the empirical observations by Hochberg (1982) con- cerning the accumulation of local geometric constraint when viewing a line drawing through a limited aperture. Consider now the 3-D interpretation of smooth curves. The approach has been to distinguish three physical categories of curve: (1) the bound- ing, or occluding, contours that comprise the silhouette boundary of a smooth shape (Marr, 1977,1982); (2) foreshortened texture contours such as the mottled pattern of light and shadow under a tree (Witkin, 1981); and (3) smooth contours that extend across a surface, such as seams, creases, wrinkles, and linear pigmentation markings (Stevens, 1981, 1986). These three, distinctly different categories of contour provide information about surface shape in different ways. Marr’s (1977) work on smooth silhouette boundaries concerns inferring the axis of an object and its shape about that axis, all on the basis of its silhouette shape in the image. Marr uses the term contour generator to refer to the 3-D locus of surface points that projects to the (silhouette) contour, and makes three geometrical assumptions about the contour generator that allow the 3-D interpretation: (1) each line of sight from the viewer to the object grazes the object’s surface at exactly one point; (2) that nearby points on the contour arise from nearby points on the contour generator; and (3) the contour generator lies wholly in a single plane. It would appear that these assumptions must be held a priori to make any concrete inferences about the shape of an object from its boundary. Interested readers are referred to Marr (1982) for continuation of this issue. What about the contours that lie interior to the silhouette of an object? Of these it is useful to distinguish between contours that follow or reflect the shape of the underlying surface (Fig. 2.1) from those whose shape is 5That is, the surface texture is random in that the contour curvature is not correlated with the tangent direction.
  • 54.
    2. VISUAL OBJECTPERCEPTION 3 3 FIG. 2.1. Contours following or reflecting shape of underlying surface. random, but whose systematic foreshortening reflects the orientation of the approximately planar surface underneath (Fig. 2.2) (Stevens, 1981; Wit- kin, 1981). With reference to Fig. 2.2, Witkin (1981) has shown that the slant and tilt of a surface patch can be estimated from a measure of the foreshortening of texture contours (there is a lawful relationship between the magnitude of the contour curvature and the orientation of the tangent along the contour), provided the viewer assumes that the texture is isotropic5 and the surface is approximately planar under the contours. When these assumptions hold, the surface orientation predicted by that computational method corresponds rather well with the apparent surface orientation. Of course, if the surface texture is actually anisotropic in a manner that mimics foreshortening, the method would fail, giving an FIG. 2.2. Foreshortening of texture contours.
  • 55.
    Random documents withunrelated content Scribd suggests to you:
  • 59.
    The Project GutenbergeBook of False Evidence
  • 60.
    This ebook isfor the use of anyone anywhere in the United States and most other parts of the world at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this ebook or online at www.gutenberg.org. If you are not located in the United States, you will have to check the laws of the country where you are located before using this eBook. Title: False Evidence Author: E. Phillips Oppenheim Illustrator: Maurice Grieffenhagen Release date: October 24, 2017 [eBook #55798] Most recently updated: October 23, 2024 Language: English Credits: Produced by Al Haines *** START OF THE PROJECT GUTENBERG EBOOK FALSE EVIDENCE ***
  • 63.
    "They sprang afterme, but started back with a quick exclamation, for they looked into the black muzzle of my father's revolver." (Chapter XXXVII.) FALSE EVIDENCE BY E. PHILLIPS OPPENHEIM Author of "Anne, the Adventuress," "The Traitors," "Conspirators," etc. WARD, LOCK & CO., LIMITED, LONDON, MELBOURNE AND TORONTO. 1911. This Book, written by the Author some years ago, is now issued in Library form for the first time.
  • 64.
    CONTENTS CHAP. PROLOGUE I. MY APOLOGY II.THE FIRST CLOUD III. "THE BOY MUST BE TOLD" IV. "A MYSTERIOUS MEETING" V. "ON BOSSINGTON HEADLAND" VI. AN INTERRUPTED ADDRESS VII. "I AM TOLD" VIII. "MY VOW" IX. AN UNEXPECTED VISIT X. THE FIRST MOVE XI. COLONEL DEVEREUX'S LAND AGENT XII. AT DEVEREUX COURT XIII. COLONEL SIR FRANCIS DEVEREUX, BART. XIV. THE BEGINNING OF DANGER XV. A FIGHT FOR LIFE XVI. MY CONVALESCENCE XVII. A MOONLIGHT RIDE XVIII. A STRANGE INTERVIEW XIX. MARIAN SURPRISES ME XX. AMONGST THE BULRUSHES XXI. RUPERT DEVEREUX XXII. FACE TO FACE XXIII. IN THE PICTURE GALLERY XXIV. A MIDNIGHT VISITOR XXV. "COUSINS!" XXVI. I "GIVE WARNING" XXVII. SIR FRANCIS DEVEREUX'S APPEAL XXVIII. GOOD-BYE TO DEVEREUX COURT
  • 65.
    XXIX. I AMTEMPTED XXX. LIAR AND COWARD XXXI. MY FATHER AND I XXXII. THE BRIGANDS' HOME XXXIII. AT PALERMO XXXIV. VISITORS FROM ROME XXXV. WE ENTERTAIN AT THE VILLA XXXVI. MR. BURTON LEIGH XXXVII. CUT DOWN XXXVIII. AN OMINOUS NOTE XXXIX. "MY FATHER'S RESOLUTION" XL. A HORRIBLE MISTAKE XLI. "TWO YEARS AFTER" XLII. A TRAITOROUS LOVE XLIII. EXPIATION XLIV. "HERO"
  • 66.
    FALSE EVIDENCE PROLOGUE The lastsally had been made and repulsed, the last shot fired; the fight was over, and victory remained with the white men. And yet, after all, was it a victory or a massacre? If you were a stay-at-home, and read the report from the telegrams in your club, or in the triumphant columns of the daily papers, especially those on the side of the Ministry, you would certainly have pronounced it the former. But if you had been there on the spot, and had seen the half-naked, ill-armed natives, with the fire of patriotism blazing in their eyes and leaping in their hearts—had seen them being shot down in rows by the merciless guns of the English batteries—another view of the matter might have presented itself to you. It might have occurred to you that these men were fighting on their own soil for their freedom and their country, and that the spirit which was blinding their eyes to the hopelessness of resistance, and urging them on to resist the stranger's progress with such passionate ineffectiveness, was after all, a natural and a poetic one. But, after all, this has nothing to do with my story. The battle was over, and it was morning. Far away in the east a dull red light had arisen from over the tops of the towering black mountains, and an angry sun was sullenly shining on the scene of carnage. It was a low hillside, once pleasant enough to look upon, but at that moment probably the most hideous sight which the whole universe could have shown. The silvery streams, which had trickled lazily down to the valley below, now ran thick and red with blood. The luxuriant shrubs and high waving ferns were trampled down and disfigured, and, most horrible sight of all, everywhere were strewn the copper-coloured forms of the beaten natives. There they lay apart and in heaps in all imaginable postures, and with all imaginable
  • 67.
    expressions on theirhard, battered faces. Some lay on their sides with their fingers locked around their spears, and the rigid frown and convulsed passion of an undying hatred branded on their numbed features. Others less brave had been shot in the back whilst flying from the death-dealing fire of the European guns, and lay stretched about in attitudes which in life would have been comical, but in death were grotesquely hideous; and over the sloping fields the misty clouds of smoke still lingered and curled upwards from the battered extinct shells which lay thick on the ground. High above the scene of devastation, on a rocky tableau at the summit of the range of hills, were pitched the tents of the victors. A little apart from these, conspicuous by the flag which floated above it, were the general's quarters; and underneath that sloping roof of canvas a strange scene was being enacted. Seated amongst a little group of the superior officers, with a heavy frown on his stern face, sat the general. Before him, at a little distance, with a soldier on either side, stood a tall, slight young man, in the uniform of an officer, but swordless. His smooth face, as yet beardless, was dyed with a deep flush, which might well be there, whether it proceeded from shame or indignation. For he was under arrest, and charged with a crime which, in a soldier, is heinous indeed—it was cowardice. It was a court-martial before which he stood arraigned, although a hastily improvised one. But soldiers have prompt ideas of justice, and General Luxton was a martinet in all matters of discipline. Disciplinarian though he was, however, he liked little the task which was now before him. He looked up from the papers, which were stretched out on the rickety little round table, with a sudden movement, and bent his frowning gaze upon the accused. The young man returned his gaze steadily, but the colour in his cheeks grew deeper. "Herbert Devereux, you stand accused of a crime which, in your profession, nothing can palliate or excuse. Have you anything to say for yourself?"
  • 68.
    "There will beno need for me to say anything, sir," was the prompt reply. "It is true that I turned my back upon the enemy, but it was to face a greater danger. The man whose life I saved can disprove this cruel charge against me in a moment. I admit that, from your point of view, appearances are suspicious, but you have only to learn from my half-brother, Rupert Devereux, why I quitted my post, and what I effected by so doing, to absolve me at least from all suspicion of cowardice, however much I may be to blame as a matter of discipline." General Luxton appeared surprised, a little relieved. "I hope so," he said, not unkindly. "Roberts, send an orderly to Lieutenant Devereux's tent, and command his presence at once." The man withdrew, and there was a few minutes' delay. Then the entrance to the tent was lifted up, and a tall, dark young man, with thin but decided features, and flashing black eyes, stepped forward. He was handsome, after a certain type, but his expression was too lifeless and supercilious to be prepossessing. General Luxton looked up and nodded. "Lieutenant Devereux, your half-brother, who stands accused of cowardice in the face of the enemy, appeals to you to give evidence on his behalf. Let us hear what you saw of him during the recent fighting." Eagerly, and with a confident light in his fair young face, the prisoner turned towards the man to whom these words were addressed. But slowly and deliberately the latter turned his back upon his half-brother without noticing his glance of appeal, and with a scornful light in his eyes. There was a slight murmur, and an interchange of looks amongst the few who were present at this significant action. "I do not know, General Luxton," he said, slowly, "what the prisoner can expect me to say likely to benefit him. He can scarcely be so mad as to expect me to shield him in this matter on account of our relationship, or to preserve the honour of our name, and yet I do not see why else he should have appealed to me. I saw very little of the affair, and would rather not
  • 69.
    have seen that.I was riding to you, sir, with a message from Colonel Elliott; and, as I passed trench 4, I saw the prisoner suddenly leave his company and run towards me. He passed several yards to the left, and as he seemed to be hurrying along aimlessly, I called to him. He made no answer, but ——" "LIAR!" The word seemed hurled out with such a passionate intensity that every one started. General Luxton looked up angrily. "Silence, sir! You will have an opportunity of saying what you have to say presently. Proceed, Devereux." "As I was saying," Rupert Devereux continued calmly, without appearing to have noticed the interruption, "he made no answer, but seemed to wish to avoid me. As the message with which I was entrusted was an important one, I rode on and left him hurrying towards the rear." With a sterner air even than he had at first assumed, General Luxton turned towards the unfortunate young man who stood before him. He was standing as though turned to stone, with wide-open eyes, staring at the man who had just spoken, attitude and expression alike bespeaking an overpowering bewilderment. "You are at liberty to ask the witness any questions," the General said, shortly. For a moment there was a dead silence. Then the words came pouring out from his quivering lips like a mountain torrent. "Rupert, what have you said? What does this mean? Good God, are you trying to ruin me? Did I not run to your assistance because you were beset by those three blackguards? Didn't I kill two of them and save your life? You can't have forgotten it! Why are you lying? Hilton saw it all, and so did Fenwick. Where are they? My God, this is horrible!"
  • 70.
    The deep flushhad gone from his cheeks, and left him pale as death. Great beads of perspiration stood out upon his forehead, and there was a wild look in his deep blue eyes. But the man to whom he made his passionate appeal kept his back turned and heeded not a word of it. Instead of answering he addressed the General. "General Luxton," Rupert said, calmly, "the accused, in denying the truth of my statement, mentions the names of two men whom he admits were witnesses of this lamentable occurrence. Might I suggest that they be called to give their version?" The General nodded assent, and the thing was done. But Hilton was the only one who answered the summons, and on reference to a list of the killed and wounded it was found that Fenwick was reported missing. "John Hilton, the accused has appealed to you to give evidence on his behalf. Let us hear what you saw of him during the recent fighting." The man, an ordinary-looking private, stepped forward and saluted. "I only saw him for a moment, sir," he said, slowly, and with a marked reluctance. "I was riding behind Lieutenant Devereux when I saw him leave his company and pass us a few yards to the left. It struck me that he looked very pale, and I thought that perhaps he was wounded." "He did not leave his company to come to your master's assistance, then?" "Certainly not, sir. We were not in any need of it. None of the enemy were near us." "Thank you. You can go, Hilton." The man saluted and went. There was a dead silence for a full minute. Then there came a passionate, hysterical cry from the prisoner—
  • 71.
    "Liar! Liar! GeneralLuxton, upon my honour, either my brother and this man are under some hallucination or they have entered into a conspiracy against me. Before God Almighty I swear that I only left my post because several of the enemy had crept down from the hill behind and had attacked my brother and his servant. I killed one of them, and the blood of the other is still on my sword. Why, Rupert, you know that you called out, 'Thanks, Herbert, you have saved my life.' Those were your very words!" The man appealed to shook his head slowly and as though with great reluctance. The sigh seemed to madden the prisoner, and he made a sudden movement forward as though to spring at him. "Oh, this is horrible!" he cried. "Where is Fenwick? He saw it all. Let him be called." General Luxton glanced again at the list before him and looked up. "You are unfortunate in your selections," he said, dryly. "The evidence of Hilton and your brother, to whom you appealed, only strengthens the case against you. Fenwick is missing. Herbert Devereux," he went on sternly, "the charge against you has been proved. I, myself, at a most critical moment, saw you desert your post when it was the centre of attack, and it fell to another's lot to lead your men on to the pursuit. The reasons which you have brought forward to account for your unwarrantable action have been clearly disposed of. You are most certainly guilty of a crime for which, amongst soldiers, there is no pardon. But you are young, and I cannot forget that you are the son of one of the most distinguished officers with whom it has been my good fortune to be associated. For his sake I am willing to make some allowance for you—on one condition you may retain your commission, and, I trust, retrieve this well-nigh fatal mistake in the future. To the crime of cowardice you have added the crime of lying; for that your account of the attack upon your half-brother and your rescue is a pure fabrication I cannot doubt. The peculiar curve in the defile behind trench 4 unfortunately hid you from the field of battle and prevents further evidence as to the occurrence which, you say, took place. But that your story is false no one can possibly doubt. The place has been carefully examined, and there are no dead bodies within a hundred yards. It seems, from your appeal to your half-brother, that you expected him to shield you at the expense of
  • 72.
    his honour. Thislie and false statement of yours you must retract if you hope for any mercy from me." There was a convulsive agony in the boy's white, strained face as he drew himself up, and looked half piteously, half indignantly at his judge. But when he tried to speak he could not, and there was a minute or two's dead silence whilst he was struggling to obtain the mastery over himself. All expected a confession, and General Luxton removed his eyes from the prisoner, and bent close over his papers, that none might read the compassion which was in his heart, and which was reflected in his face. The words came at last; and shrill and incoherent though they were, there was a ring of genuine dignity in them. "General Luxton, I have been guilty neither of cowardice nor falsehood. I swear before God, on the sword which my father himself put into my hands before I left England; by everything that is most holy to me I swear that my account of this awful occurrence is true. Ask the men of whom I was in command when I caught sight of—of him"—and he pointed with a trembling finger and a gesture than which nothing could have been more dramatic to his half-brother—"ask them whether I bore myself like a coward when those spears were whistling around us, or when we were fighting hand-to-hand after the first repulse. God knows that I did not. I left my post to encounter a greater danger still. Bitterly do I regret that I ever did so; but it is the only indiscretion of which I am guilty. I swear it." General Luxton raised his head, and what there had been of compassion in his face was either gone or effectually concealed. "You have sworn enough already," he said, sternly. "Herbert Devereux, I am bitterly disappointed in you. I was willing to spare your father the disgrace which I fear will kill him; but you cut away the ground from under my feet. You are most certainly proved guilty of gross cowardice in the face of the enemy found guilty, not upon the evidence of one man, but of two, and one of those your own relative. Circumstances, too, are strong against you, so are the probabilities. Most undeniably and conclusively you are found guilty; guilty of cowardice, guilty of falsehood. You will remain under arrest until I can find an opportunity of sending an escort with you to
  • 73.
    the Hekla. Yourcommission is forfeited to the Queen, whose uniform you have disgraced." Never a sign of guilt in the prisoner's countenance. Proudly and indignantly he looked his General straight in the face, his cheeks red with a flush, which was not of shame, and the wild fury in his heart blazing out of his eyes. "It is not I who have disgraced the Queen's colours; but he—he who has fabricated and sworn to a false string of lies. Rupert, in your heart alone is the knowledge of why you have done this thing. But some day you shall tell me—or die." There was something intensely dramatic in the passionate bitterness which vibrated in the shrill boyish tone, and, as though moved by a common impulse, every one in the tent followed that threatening gesture. But the face of Rupert Devereux was little like the face of a guilty man. He looked somewhat agitated, and a good deal pained; but although he was the cynosure of all eyes, he turned never a shade the paler, nor flinched once from the passionate fire which was leaping from the eyes of the young prisoner. He seemed as though about to make some reply; but the General raised his hand. "Remove the prisoner." There was a sudden commotion, for, with a deep, despairing groan, and arms for a moment lifted high above his head, he had staggered backwards and sunk heavily to the ground in a dead swoon. What wonder! He was but a boy after all. * * * * * "Herbert! Why, Herbert! Good God! where did you spring from? Are you invalided?" The moonlight was streaming in through the high oriel windows of the long picture-gallery, glittering upon the armour and crossed weapons which hung upon the walls, and casting fantastic rays down the polished oak floor.
  • 74.
    Colonel Sir FrancisDevereux dropped the cigar which he had been peacefully smoking, and brought to a sudden halt his leisurely perambulation of this his favourite resort. Before him, with drooping head, with sunken cheeks, and with deep black rims under his eyes, stood his son Herbert, who, only a few months ago, had departed on his first campaign, a happy, careless young sub. Was it, indeed, his son, or was it a ghost that had stolen upon him out of the gloomy shadows of the vast gallery? "Invalided! Would to God that I was dead!" broke from the boy's quivering lips. "Father, I have brought disgrace upon you—disgrace upon our name." And he stretched out his hands towards the long line of pictured warriors, who seemed to be frowning down upon him from the wall. "Disgrace that you will never forgive, never pardon." Like a statue of stone the proud old soldier stood while he listened to his son's story. Then, with a half-smothered groan, he deliberately turned his back upon him. "Father," he pleaded, "listen to me. Before heaven I swear that I am innocent. Rupert lied. Why, I don't know, but he lied. I never felt fear." His father turned half round. "You have been put on your defence. General Luxton would never have found your father's son guilty of cowardice had there been room for doubt. The charge was proved against you in court-martial." "But, father, it was because they believed Rupert and his man. The only two other men who saw the struggle are dead." Colonel Devereux turned away and buried his face in his hands. "A Devereux guilty of cowardice!" he groaned. "My God! that it should have been my son!" Then with a sudden movement he turned round. His son had sunk upon his knees before him, and the moon was throwing a ghastly light upon his haggard, supplicating face.
  • 75.
    "Out of mysight, and out of my heart for ever, Herbert Devereux!" cried his father, his tones vibrating with a passionate contempt. "You have brought disgrace upon a stainless name. Curse you for it, though you be a thousand times my son. You shall not sleep under this roof again. Begone! Change your name, I command you! Forget that you are a Devereux, as I most surely shall. Turn linen-draper, or man-milliner, or lawyer, what you will so that I never see or hear from you again. Begone, and curse you." Scathing and vibrating with scorn though the words were, they seemed to touch a chord in the boy's heart, not of humiliation, but of righteous anger. He sprang to his feet, and held himself for a moment as proudly as any of his armoured ancestors who looked down from the walls upon father and son. "I will go, then," he cried, firmly. "It is right that I should go. But, after all, it is false to say that I have disgraced your name. It is Rupert who has done this." He turned and walked steadily away, without a backward glance. Out of the swing doors on to the broad staircase, he passed along noble corridors, between rows of marble statues, down into the mighty dome-like hall, and out of the house which he had loved so well. And the servants, who would have pressed forward to welcome him, hung back in fear, for there was that in his face which they shrunk from looking upon. Out into the soft summer night he stepped, heedless of their wondering glances, and down the broad avenue he hurried, never pausing once to breathe in the balmy night wind, heavy with the odour of sweet-smelling flowers, or to listen to the nightingale singing in the low copse which bordered the gardens. Through a low iron gate he stepped into the park, and walked swiftly along, never glancing to the right or to the left at the strange shadows cast by the mighty oak-trees on the velvety turf, or at the startled deer, who sprung up on every side of him and bounded gracefully away, or at the rabbits who were scampering about all around in desperate alarm; once he had loved to watch and to listen to all these things; but now he felt only a burning desire to escape from them, and to find himself outside the confines of the home which he was leaving for ever. And not until he had reached the last paling, and had vaulted into the broad, white road, did his strength desert him.
  • 76.
    Then, faint andweary, and heartsick, he sank down in a heap on the roadside, and prayed that he might die. * * * * * A cloudless summer morning, with the freshness of dawn still lingering in the air. A morning which seemed about to herald in one of Nature's perfect days, on which to be sad were a crime, and to have troubles absurd. Already the dreamy humming of bees was floating in the atmosphere, and the lark had given place to noisier, if less musical, songsters. It was a glorious morning. Over the low, iron gate of an old-fashioned garden a girl was leaning, her head resting lightly upon her hand, gazing across the pleasant meadows to the dark woods beyond, with a soft, far-away look in her grey eyes—for she was thinking of her lover. She was dressed in a blue print gown, which hung in simple folds around her straight, slim figure, and she had carelessly passed the long stalk of a full-blown red rose within her waistband. It was a very pleasant view that she was admiring; but any casual spectator would have declared that she was the most charming object in it. And there was a spectator, although not a casual one. Suddenly, like a ghost, the figure of her dreams stood before her. Pale, haggard, and dishevelled-looking, he seemed to have risen out of the very ground; and it was very little to be wondered at that, at first, she shrunk back alarmed. "Herbert! Herbert! can it really be you?" He never answered her; but, as the first surprise began to fade away, she moved forward, and would have thrown herself into his arms. But he stopped her. "Keep back, Marian," he cried, hoarsely; "keep away from me! I have come to bid you good-bye." A swift, sudden fear drove the colour from her cheeks, and chilled her through and through; but she faltered out an answer.
  • 77.
    "Good-bye, Herbert! Whatdo you mean? Oh, tell me what has happened, quick!" "The one thing worse than death, Marian—disgrace!" And then, with his face turned away, and his eyes resting wearily on the picturesque landscape, he told her his story. * * * * * The last word had left his quivering lips, and he stood as though in a dream. The worst was over. He had told his father, and he had told her. It seemed like the end of all things to him. Suddenly a pair of white arms were thrown around his neck, and a great red rose was crushed to pieces against his waistcoat. "Herbert! oh, Herbert! how dreadful! Don't look like that, you frighten me!" He was striving to free himself, but she would not let him go. "Dearest, you don't understand! This is ruin to me. My father has turned me from the house, commanded me to bear another name, disowned me. Be brave, Marian, for we must part. I am here only to tell you this, and to bid you farewell." Still she would not let him go. "You will do nothing of the sort, sir. I'll not be thrown over in that fashion," she said, struggling to smile through her tears. "And, Herbert, oh, Herbert! how ill you look! You've been out all night." He did not deny it, but again he strove to disengage himself. But she would have none of it. "Bertie, dearest," she spoke cheerfully, though her eyes were still swimming with tears, "you mustn't think that you're going to get rid of us in
  • 78.
    this way. You'vejust got to come in to breakfast with me, and afterwards we'll tell Grannie all about it. Come along, sir, I insist." He braced himself up for resistance, but he had still to learn that against a woman's love a man's will can prevail nothing. At first he was firm, then wavering, and finally he was led in triumph across the smooth lawn and along the winding path to the French windows of the morning-room. But when he found himself face to face with the kind old lady who had loved him as her own son, and saw the tears trickle down her withered, apple-red cheeks as she listened to the tale which Marian poured out, he felt that he had passed the limits of self-endurance. For more than twenty-four hours he had neither eaten nor drunk, and he was sick at heart. Gradually Marian felt the arm, which she had drawn tightly through hers, grow heavier and heavier until at last as she finished her tale with a little tremulous burst of indignation, he sank back in the arm-chair, and slowly fainted. But through the mist which closed in upon him he saw nothing but kindly pitying faces bending over his, and heard Grannie's gentle whisper— "I believe you, Herbert," and more emphatic but none the less earnest were her words, whose sweet, tear-stained face, so close to his, was the last he saw when unconsciousness was closing in upon him. "So do I, Bertie, I hate Rupert," and sweeter than the most heart-stirring music were the faltering words she added— "And I love you better than ever. Oh, Grannie, Grannie, he has fainted!" CHAPTER I MY APOLOGY Fortune is the strangest mistress a man ever wooed. Who courts her she shuns, who deserves her she passes over, and on him who defies her and takes no pains to secure her she lavishes her favours. I am one of those to
  • 79.
    whom she hasshown herself most kind. Many years ago I vowed my life away to one purpose, and that partly an immoral one. It was a purpose which held my life. I swore to seek no end apart from it, and I put away from my thoughts all joys that were not included in its accomplishment. And yet, having kept my oath, I still possess in the prime of life everything which a man could wish for. I am rich, and well thought of amongst my fellows. I am married to the woman whom I love, and life is flowing on with me as calmly and peacefully as the murmuring waters of a woodland stream in the middle of summer. And, above all, my heart is at ease, for I have kept my vow. She is a strange mistress, indeed! Nothing have I sought or deserved of her, yet everything I have. Whilst he who was far above me in his deservings, and whose sufferings none save myself thoroughly understood, passed through a gloomy life, buffeted by every wind, stranded by every tide of fortune; misunderstood, wronged, falsely accused, and narrowly escaped remaining in men's minds only as a prototype of a passionate, unforgiving, Quixotic man. That the world may know him as he was, and form a better judgment as to his character, I have gathered together the threads of my life indissolubly connected with his, and have turned them inside out. I have never indulged myself with the feminine luxury of a diary, but with a surer progress than of pen over paper has the record of my strange life been written into my mind; and so I tell it just as it all comes back to me, not as a professed story-teller, with harmonious dates and regular evolution of plot, and neatly paged chapters, but in a bolder way, leaving much to be guessed at, and some things untold. If there be any of whom I have occasion to speak still amongst the living (my life has so contracted of late that many have passed out of its horizon), let them remember for what purpose I write, and for his sake forbear to complain. If the sword were the pen, then would mine be the pen of a ready writer, and I might be able to touch lightly on their shortcomings, and gild over the black spots on my own life. But enough of excuses. I take up my pen a blunt Englishman, an athlete rather than a scholar, to write a plain story which shall serve not as a eulogy, but as a justification of the man to whom many years of my life have been
  • 80.
    ungrudgingly given. Letall those who may feel disposed to cavil at the disconnectedness of my loosely jointed story, remember this, and be silent. CHAPTER II THE FIRST CLOUD About a mile seaward from Porlock, separated from it by a narrow strip of the most luxuriant meadowland in Devonshire, lies the village of Bossington. Perhaps it were better called a hamlet, for at the time when I knew anything about it (which, let the tourist remember, is many years ago) it consisted but of six or seven cottages, a farmhouse, and a half-ruined old manor-house, for the privilege of living in which my father paid ten pounds a year, or some such trifling sum, to the neighbouring clergyman whose property it was. But what the place lacked in size was certainly atoned for—and more than atoned for—by the beauty of its situation. High above it, like a mighty protecting giant, rose Bossington Headland, covered always with a soft, springy turf, and glowing in midsummer with the brilliant colouring of rich purple heather and yellow gorse. Often have I stood on its highest point, and with my head bared to the strong fresh breeze, watched the sun rise over the Exmoor Hills and Dunkerry Beacon, and waited until it shed its first warm gleams on the white cottages and queer old church-tower of Porlock, which lay clustered together in picturesque irregularity at the head of the little bay. And almost as often have I gazed upon the same scene from the same spot by the less distinct but more harmonious light of the full harvest moon, and have wondered in which guise it seemed the fairest. Behind Bossington lay Allercombe Woods, great tree-covered hills sloping on one side down to the road which connected, and still connects, Porlock with Minehead and the outside world, and on the other, descending precipitously to the sea; so precipitously indeed that it seemed always a wonder to me how the thickly growing but stunted fir-trees could preserve
  • 81.
    their shape andregularity. The descent from Bossington Headland into Porlock was by a steep winding path through Allercombe Woods, and many a time I have looked through the thin coating of green leaves upon the fields which stretched like a piece of patchwork below down to the sea, and wondered whether any other country in the world (I had never been out of Devonshire then) could be more beautiful than this. Within a stone's throw of where the blue sea of our English Bay of Naples rippled in on to the firm white sands, was the tumble-down old building in which we lived. What there had been of walls had long before our time been hidden by climbing plants and ivy, and in summer-time the place from a distance somewhat resembled a gigantic nosegay of cottage roses, jessamine, and other creeping flowers. There was but a small garden and no ground, for Bossington Headland rose precipitously close to the back of the house, and in front there was no space for any. A shed served as a stable for one or two Exmoor ponies, and also as a sleeping-place for the lanky, raw-boned Devonshire lad whom we kept to look after them. There were but few habitable rooms in our mansion, but they were sufficient, for our household was a small one. My father, mother, sister, myself, and a country servant comprised it. We never had a visitor, save occasionally the clergyman from Porlock. We never went anywhere. We knew no one, and at seventeen years of age an idea which had been developing in me for a long time, took to itself the tangible shape of words. "Father," I said to him one evening when we were sitting out upon our little strip of lawn together, he smoking, I envying him for being able to smoke, "do you know that I have never been out of Devonshire—never been further than Exeter even, and I am eighteen years old?" It was long before he answered me, and when, at last, he turned round and did so, I was distressed to see the look of deep anxiety in his worn, handsome face, and the troubled light in his clear eyes. "I know it, my boy," he said, pityingly. "I have been expecting this. You are weary of the country."
  • 82.
    I stood up,with my hands in my pockets, and my back against the latticed wall of the house, gazing over the sparkling, dancing sea, to where, on the horizon, the stars seemed to stoop and meet it. Was I tired of this quiet home? I scarcely knew; country sports and country sights were dear to me, and I had no desire to leave them for ever. I thought of the fat trout in the Exford streams, and the huntsman's rallying call from "t'other side Dunkerry," and the wild birds that needed so much getting at and such quick firing, and of the deep-sea fishing, and the shooting up the coombes from Farmer Pulsford's boat, and of the delight of shipping on a hot summer's day and diving deep down into the cool bracing water. Why should I wish to leave all this? What should I be likely to find pleasanter in the world of which, as yet, I knew nothing? For a moment or two I hesitated —but it was only for a moment or two. The restlessness which had been growing up within me for years was built upon a solid foundation, and would not be silenced. "No, I'm not tired of the country, father," I answered, slowly. "I love it too much ever to be tired of it. But men don't generally live all their lives in one place, do they, without having any work or anything to do except enjoy themselves?" "And what should you like to be?" my father asked, quickly. I had long ago made up my mind upon that point, and was not slow to answer— "I should like to be a soldier," I declared, emphatically. I was very little prepared for the result of my words. A spasm of what seemed to be the most acute pain passed across my father's face, and he covered it for a moment with his hands. When he withdrew them he looked like a ghost, deathly pale in the golden moonlight, and when he spoke his voice trembled with emotion. "God forbid that you should wish it seriously!" he said, "for it is the one thing which you can never be!"
  • 83.
    "Oh, Hugh, youdo not mean it really; you do not wish to go away from us!" I turned round, for the voice, a soft and gentle one, was my mother's. She was standing in the open window with a fleecy white shawl around her head, and her eyes, the sweetest I ever saw, fixed appealingly upon me. I glanced from one to the other blankly, for my disappointment was great. Then, like a flash, a sudden conviction laid hold of me. There was some great and mysterious reason why we had lived so long apart from the world. CHAPTER III "THE BOY MUST BE TOLD" That was quite an eventful night in our quiet life. Whilst we three stood looking at one another half fearfully—I full of this strange, new idea which had just occurred to me—we heard the latch of our garden gate lifted, and Mr. Cox, the vicar of Porlock and my instructor in the classics, followed by no fewer than four large-limbed, broad-shouldered, Porlock men, entered. They made their way up the steep garden path, and my father, in no little surprise, rose to greet them. With Mr. Cox he shook hands and then glanced inquiringly at his followers, who, after touching their hats respectfully, stood in a row looking supremely uncomfortable, and each betraying a strong disposition to retire a little behind the others. Mr. Cox proceeded to explain matters. "You are pleased to look upon us as a deputation," he said, pleasantly, waving his hand towards the others, "of which I am the spokesman. We come from the Porlock Working Men's Conservative Club." My father bowed, and bidding me bring forward a garden seat, requested the deputation to be seated. Then he called into the house for Jane to bring out some jugs of cider and glasses, and a decided smile appeared on the
  • 84.
    somewhat wooden facesof the deputation. I was vastly interested, and not a little curious. When the cider had been brought and distributed, and a raid made upon the tobacco jar, Mr. Cox proceeded with his explanation. "We have come to ask you a favour, Mr. Arbuthnot," he said. "We are going to hold a political meeting in the school-room at Porlock next week. A gentleman from Minehead is going to give us an address on the land question which promises to be very interesting, and Mr. Bowles here has kindly promised to say a few words." The end man on the seat here twirled his hat, and, being nudged by his neighbour, betrayed his personality by a broad grin. Finally, to relieve his modesty, he buried his face in the mug of cider which stood by his side. "The difficulty we are in is this," continued Mr. Cox; "we want a chairman. I have most unfortunately promised to be in Exeter on that day and shall not be able to return in time for the meeting, or else we would not have troubled you. But as I shall not be available, we thought that perhaps you might be induced to accept the office. That is what we have come to ask you." My father shook his head. "It is very kind of you to think of me," he said, hesitatingly, "but I fear that I must decline your offer. Politics have lost most of their interest for me —and—and, in short, I think I would rather not." "I hope you will reconsider that," Mr. Cox said, pleasantly. "It will be a very slight tax upon you after all. You need only say a very few words. Come, think it over again. We really are at our wit's end or we would not have troubled you. "There is Mr. Sothern," my father protested. "He is in bed ill. An attack of pleurisy, I think."
  • 85.
    "Mr. Brown, then?" "Arank Radical." "Mr. Jephcote?" "Away." "Mr. Hetton?" "Gone to London for a week." "Mr. Smith, then?" "Will be at Exeter cattle fair." My father was silent for a moment or two. Then he suggested some more names, to each of which there was some objection. "You do seem to have been unfortunate," he declared, at last. "To tell you the truth, Mr. Cox," he added, thoughtfully, "I scarcely know what to say. I had made up my mind, for certain private reasons, never to have anything to do with public life in any shape or form." "This isn't a very formidable undertaking, is it?" Mr. Cox urged, smiling. "It isn't. But the principle is the same," my father answered. "However, leave it in this way if you like. Give me until to-morrow evening to think the matter over, and in the meantime see if you can't find some one else. I'm afraid I can't say more than that." The deputation thought that nothing could be fairer than this, and nothing more satisfactory except an unqualified assent. I think my father imagined that having promised so much they would take their departure. But nothing of the sort happened. Perhaps they found the cider too good, or perhaps they were tired after their day's work and the walk from Porlock. At any rate, there they sat for more than an hour, taking occasional gulps at their cider, and puffing incessantly at their blackened pipes with a stolid vacuous look on their honest faces, whilst my father and Mr. Cox talked a
  • 86.
    little aside ina low tone. I fancied that I was the subject of their conversation, but though I strained my ears in the attempt to catch some part of it, I was unsuccessful. Once or twice the sound of my name reached me, but directly I leaned forward they dropped their voices, so that I could hear no more. I have always believed, however, that my father was asking advice from Mr. Cox concerning me, and that Mr. Cox was urging him to send me to the University. But I never knew for certain, for events were soon to occur which swept out of my mind all minor curiosity. At last Mr. Cox rose to go, and the deputation, with manifest reluctance, did the same. My father courteously accompanied them to the garden gate, and shook hands with them all, thanking them for their visit. When he returned there was a slight sparkle in his eyes, and an amused smile on his lips. So monotonous was our life, that even such an event as this was welcome, and I could tell from his manner that he was pleased at the request which had been made to him, and disposed to accept it. I determined to encourage him in it. "Governor," I remarked, leaning over the wall and watching the retreating forms of our visitors, "I hope we're not going to have many political deputations here, especially if they're all going to be as thirsty as this one was. Did you ever see such fellows for cider! We shan't have a drop left for the hot weather if you encourage this sort of thing. But you'll do what they want you to, won't you? I should! It'll be capital fun, and I'm sure you'd make a rattling speech. You're up on the land question, too. I heard you giving it to old Simpson the other morning." My father smiled, and stood by my side watching them make their way down the coombe. "I shall have to consult your mother about it," he said. "I almost think that I may venture it," he added, in a lower tone and thoughtfully, as though to himself. "Venture it! What could there be adventurous in it," I wondered, "to a well-read, scholarly man such as I knew him to be!" But I did not dare to ask.
  • 87.
    Presently he turnedto me with a much graver look in his face. "Hugh!" he said, "these people interrupted our conversation. There is something which I must say to you at once. I do not wish you to become a soldier. When you feel that you can stay here no longer, and that this country life is too quiet for you, you must choose some other profession. But a soldier you can never be." I was bitterly disappointed, and not a little curious, and an idea which had often occurred to me swept suddenly into my mind with renewed strength. "Father, may I ask you a question?" He hesitated, but did not forbid me. "I have heard it said down in the village—every one says that you must once have been a soldier. You walk and hold your head like one, and— father, what is the matter?" I broke off all at once, for his face had become like a dead man's, and he had sunk heavily on to the seat. I would have sprung to his side, but my mother was there before him. She had passed one arm around his neck, and with the other she motioned me to go into the house. "It isn't your fault, Hugh," she said, "but you mustn't ask your father questions; they distress him. Leave us now." I turned heavily away, and went up-stairs to my room. About an hour afterwards, when I pushed open my window before getting into bed, there stole into my room together with the sweet scent of jessamine and climbing roses the sound of subdued voices. "He must be told," I heard my father say solemnly. "God give me strength." Then the voices ceased for a while, but I still lingered, and presently they began again, but in a more cheerful key.
  • 88.
    Welcome to ourwebsite – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com