1. 518
J.Soc.Photogr.Sci.Technol.Japn. (2005)Vol.68No.6: 518-531
一般論文
"How We Look at Photographs"
As Indicated by the Ability to Discriminate Contrast in Grey-Scale vs. Photographs
コン トラス ト感度 に基 づ く写 真の見方の研究
Sharon GERSHONI
*
and Hiroyuki KOBAYASHI
*
Sharon GERSHONI
*
・小 林 裕 幸
*
Abstract In order to examine the roles and interferences of local and global elements in lightness perception and object recognition
processes of photographs with meaningful contents, we examined whether contrast discrimination is a response to spatial
configuration properties of photographs, or also a function of conceptual contents. In three experiments we compared contrast
discrimination performances of observers, when presented with contrast increments applied to discrete tonal regions in grey-
scales and in several categories of black-and-white photographs of natural scenes. In Experiment 1, observers performed
contrast discrimination in grey-scales by rank-order tasks. In Experiments 2 and 3, trained and novice observers performed
contrast discrimination of photographs by sortingtasks. We found substantial differences in response to contrast increments,
depending on the region, but no significant effect of category. Nevertheless, low performances in shadow region of grey-scales,
significantly improved in photographs due to complexity of configuration. We also found differences in performance between
photographs of light and night scene.
要 旨 本研究は 白黒写真 の輝度対比が もた らす美学的経験 と,輝 度対比を検知 あるい は感ず る能 力,す なわち観察者 のコン トラ
ス ト感度 との関係を調べた ものである.コ ン トラス ト感度が,写 真 の物理特性(テ ー マのない刺激)に だ け依存す るのか,
それ とも写 真 の絵柄 やテー マ(テ ー マのあ る刺激)に も依 存す るのかを明 らかにす るため に,テ ーマ のある刺 激 と して
Ansel Adamsの 写真を,そ してテー マのない刺激 としてグ レースケール画像 を用 い,そ れぞれの コン トラス ト感度 を心理
学的尺度構成法 によ り比較 した.Ansel Adamsは ゾー ンシステムを用 い,メ リハ リがあ り,階 調豊か な写真 を作 った写 真
家で あるが,Adamsが 見,そ して感 じた ものが,ロ ー カルコン トラス トを調節す る ことで強調 され てい ること,そ して
ロー カル コン トラス トは アンカーの よ うにローカル フレー ム ワー クを強調 し,高 い明度恒常性 のあ る観察条 件を もた ら
し,美 的感覚 を増大 させ る ことがわ かった.
Key words: contrast discrimination, lightness perception, black-and-white photography
キ ー ワ ― ド:コ ン ト ラ ス ト感 度,白 黒 写 真 と グ レ イ ス ケ ー ル の 比 較,概 念 的 要 素 と 空 間 配 置 要 素 の 比 較,Anchoring理 論,ロ ー カ
ル フ レー ム ワ ー ク と グ ロ ー バ ル フ レ ー ム ワ ー ク の 比 較
1. Introduction
In recent years there is a growing interest in the correlation
between the physical structure of pictures and perceptual and
cognitive functions. Furthermore, analysis of the process of
making pictures offers new approaches to understanding the
measured physical properties as technical means of the aesthetic
expression as the products of the physiology of the brain and its
constraints. A theory of aesthetics as a biological function has
shown the assimilation between art and brain based on simi-
larities in their aims, strategies and functions (Zeki, 2000) 1).
Another theory proposed laws of artistic experience, based on
neurobiological strategies with high survival value, such as
the peak shift effect and gestalt grouping principles, as a brain
response-mechanism model that applies to every work of art
(Ramachandran and Hirstein, 1999) 2).
Though until recently, researched as been focused mainly on
paintings, and overlooked photography as a means for investi-
gation of the mechanisms mentioned above, the presented
research suggests photography as an appropriate case, for the
following reasons:
(1) Photography has roots in both the tradition of art and repro-
duction technologies, and its main concern is the relation-
ship between properties of light sensitive material and the
visual phenomena (perceptual and cognitive).
(2) As a medium of expression, photography is composed of the
interaction between two basic motivations: the synthetic
approach that is the creation of a visual statement about the
Received 30th, September 2005, Revised and accepted 7th, October 2005
平 成17年9月30日 受 付 平 成17年10月7日 改 訂 受 付 ・受 理
*
Department of Information Science, Graduate School of Science and Technology, Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba 263-8522, Japan
千 葉 大 学 工 学 部 〒263-8522千 葉 市 稲 毛 区 弥 生 町1-33

2. S. GERSHONI and H. KOBAYASHI " How We Look at Photographs" 519
subject, and an analytical approach- the direct representa-
tion of the subject 3).
(3) The argument of transferability thesis 4) i.e. conditions for
"pictorial sameness"
, under which a reproduction is
aesthetically valuable as the original, and the opposing
claim, that pictorial sameness of a reproduction cannot
result in sameness of aesthetic value 5), are solved for
photography, since photography itself is a reproduction
system.
Black-and-white photographs in particular are an interesting
case, since they represent subject properties as relationships
between correlating luminance values, with only a one-
dimensional luminance grey-scale or luminance contrast for
input to the lightness perception processes, and the interacting
higher processes: object recognition, grouping and segmenta-
tion, etc. discussion follows.
Earlier approaches to perception of lightness assumed low-
level processing: the adaptation level theory (Helson, 1962,
1964) 6)7),presumed the mean luminance of the visual field to be
the reference and that the visual system responds to changes
above that average. The main weakness of the theory was that
in a visual field, composed of a test object and surrounding
uniform backgrounds, background alone would determine the
mean luminance hence it fails to represent real world conditions.
Simultaneous contrast theory (Wallach, 1948) 8) suggested a
comparison of the amount of light reflected by the object with
that of adjacent objects. The strength of the theory was that it
solved the dependency of object luminance on surrounding area
size (Leibowitz, 1965) 9),and its weakness that different spatial
configurations result in different perceptual organizations, which
affect the simultaneous contrasts. Ratio rule theory, based on
the physiological lateral antagonism mechanism, derived the
lightness of an object from the ratio of its luminance to that of
its surround. It failed in cases of large differences between
luminance of object and its surround (Jameson and Hurvich,
1964)10). Although Retinex model (Land and McCann, 1971) 11),
by integrating edge information over space managed to recon-
struct reflectance of complex stimuli such as 'Mondreans', it did
not suffice for real world objects, affects by configuration and
meaningful interpretations (Knill and Kersten, 1991) 12).After
the effect of interpretation given to pattern over perception of
lightness was shown (Gilchrist, 1977) 13),it became evident that
lightness perception involves three level processes, i.e. low,
high and a mid-level, where surface, contour and grouping are
processed. This approach was based on the gestalt perceptual
organization and mid-level mechanisms. More recently, Intrinsic-
image theory suggested that the luminance image is constructed
by its reflectance and illuminance (Barrow and Tenenbaum,
1978; Mend, 1994; Adelson and Petland, 1996) 14-16).Most
compatible with gestalt emphasis on the role of configuration
and perceptual organization (higher-level processes) is the
Anchoring Theory (Gilchrist and Cataliotti, 1994; Gilchrist, A.,
and Bonato, F, 1995) 17)18).The anchoring model was initiated
with the highestluminance rule, according to which, in mapping
luminance into a lightness scale, the highest luminance is
anchored (assigned) to white, and the rest of the values are
scaled relative to it (also Brown, 1994; McCann, 1994; Schirillo
and Shevell, 1996) 19-21).Other factors found to influence
anchoring are the area rule, Tjunction and configural attributes
of the image, such as articulation, insulation and gestalt grouping
principles; the result is a compromise between the rules that
determines the anchor. The anchor can occur within a local
framework, containing a group of patches or a global framework
that could include even the entire visual field. Earlier experi-
ments in object recognition showed that grouping is not prior
to lightness constancy, but a later process, composed of both
low and high processing (Beck, 1975, Olson and Atteneave,
1970)22)23),and even influenced by luminance constancy (Rock,
Nijhawan, Palmer and Tudor, 1992) 24).In addition, it was also
shown that grouping is based on similarity of complete shapes
(Palmer, Neff and Beck, 1996)25).Gestalt approach insists per-
ceptual grouping, such as region segmentation, to be a global
and direct process independent of prior processes such as
edge-detection (Shi and Malik, 1997) 26),and furthermore, that
previous experience, viewing of the object's stable properties,
causes a future perceptual organization and object recognition
to involve attention-based selection (Wertheimer, 1924)27),such
as figure/ground (Peterson and Gibson, 1991, 1993; Peterson,
1994)28-30).The crucial effect of figural familiarity on the relative
role of low and high level processes of grouping is further sup-
ported by phenomena like visual completion and amodal com-
pletion, and the hypothesis of nonaccidentalness, predicting that
the visual system prefers to process information arising from
accidental viewpoint, if they violate the structural regularities
of the general viewpoint (Rock, 1983; Lowe, 1985; Albert and
Hoffman, 1995) 31-33).In addition, it was found that 40% of V2
cells fired when presented with illusory contours (Peterhans
and von der Heydt, 1989, 1991) 34-35).Several contradicting
findings are that there is a process occurring on both sides of
the contour prior to figure/ground recognition (Peterson and
Gibson, 1991, 1993, 1994; Vecera and O'Reilly, 1998) 28-30)36)
and that internal details of object features are necessary for
matching with the object memory (Mitsumatsu and Yokosawa,
2002) 37). Other approaches to part/whole hierarchy are: a
boundary based region segmentation based on detection of
luminance at edges (Marr and Hildreth, 1980) 38),and the uni-
form connectedness, as a principle for the partition of the visual
field into units of uniform spatial characteristics (Palmer and
Rock, 1994a, 1994b) 39)40)or connected regions that was exam-
ined successfully with textured pictures Thompson, Chronicle
and Collins, 2003) 41).
In summery, either information processing proceeds from
local to global level, with a feedback from global level to facilitate
processing of local elements, or global wholes allow conscious
access to early processing results. In any case, both local and
global processes govern lightness perception and object rec-
ognition alike, in what seems to be mutual interference.
The purpose of the present study of lightness perception in
photographs lies in the fact that photography is a language of
tones and contrast relationships functioning as letters and
words to communicate impressions and delivers aesthetic
experiences, and which all its vocabulary is composed solely by
quantitatively measurable luminance values, hue, chroma and
contrast. Moreover, there is already extensive understanding in
lightness processing of simple images, or complex images such
as scales, or Mondrian patterns 42)43),that can be examined for
viewing photographs as aesthetic images with meaningful con-
tents.
To gain understanding of the roles and interferences of local
and global elements in lightness perception and object recogni-

3. 520 J.Soc.Photogr.Sci. technol.Japan Vol.68 No.66 (2005)
tion processes of meaningful images, we asked whether contrast
discrimination for meaningless images such as grey-scales
differs from photographs, and if so in what ways? and whether
contrast discrimination in photographs is only a response to
spatial configuration properties, or also a function of conceptual
contents, such as category? In three experiments we compared
contrast discrimination performance of observers, when pre-
sented with contrast changes, applied to discrete tonal regions in
grey-scales and in several categories of black-and-white photo-
graphs of natural scenes.
1.1 A Note On The Stimuli
The stimuli were photographs of Ansel Adams 44)(USA 1902-
1984), since they stand out as one of the peaks in the history
of photography, manifesting both synthetical and analytical
approaches, and are a pioneering conscious attempt to combine
them methodologically into one whole expressive framework.
Adams belonged to group "f/64" following Edward Weston's
Objectivism, which opposed subjectivity in pictorial aesthetics
declaring photography's ultimate goal to be a scientific expres-
sion of esthetics, "the absolute impersonal recognition of the
significance offacts" 45)."f/64" manifesto from 1932 demanded
straight photography to draw solely upon its inherent technical
qualities (i.e. "the microscopic revelation of the lens") and be
defined as an art-form by only a simple and direct presentation
through purely photographic methods 10).In order to achieve
such representation of subject matters with a wide variety of
luminance ranges, Adams developed the fundamental concepts
ofphotographic visualization and the Zone System. Zone System
is based on nineteenth century sensitometry of Harter and
Driffield (D-log-E Curve in 1890), and a grey-scale of ten light
intensities, corresponding to a tone-scale from white to black in
the print 46).The system offered maximum range of tones and
contributed to the tonal and contrast control, and enabled the
treatment of discrete subject areas of the image for tonal
emphasis. Photographic "Visualization" is a process for envi-
sioning the final print 'tonal management', step-by-step from
camera adjustments, through film development and printing.
Adams' photographic visualization and Zone System are equiva-
lents to the scaling and anchoring in intrinsic image 47)light-
ness perception models (see above). Furthermore, Adam's
printing method for the expression of emotional experiences is
not the mere recreation of the subject, but its enhancement by
means of local contrast and extraction of features, to grab atten-
tion to local regions of change, where processing is "more inter-
esting or pleasing" 2).
2. Stimuli Preparation
2.1 Grey-Scales Rank Order Task (Experiment 1)
Kodak (Scientific Imaging Systems) Gray Scale, consisting of
20 steps with 0.10 density increments from nominal "white" of
0.05 to 1.95 and an 18% neutral grey background. Sample size
was 20 cmx3 cm, with a 3 cm wide neutral grey masking.
2.2 Photograph- Sorting Task (Experiments 2 and 3)
Nine black-and-white photographs by the photographer Ansel
Adams 44)were used as stimuli for the present study:
1. "Spanish Peaks, Colorado", 1951, page 18
2. "Canyon de Chelly,National Monument, Arizona", 1947, page
30
3. "Moonrise, Hernandez, New-Mexico", 1944, page 55
4. "NavajoWoman, Wide Ruin, Arizona", 1948, page 39
5. "Maynard Dixon, painter, Tucson, Arizona", 1944, page 16
6. "Martha Porter, pioneer woman, Ordeville, Utah", 1961, page
81
7. "AdobeDwellings, Northern New-Mexico", 1958, page 49
8. "Santuario de Chimayol, New-Mexico", 1950, page 50
9. "Arches, North Court, Mission San Xavier del Bac, Tucson,
Arizona", 1968, page 94
The photographs belong to three major themes in photography:
Landscape (photographs no. 1, 2 and 3), Portrait (no. 4, 5 and 6)
and Architecture (no. 7, 8 and 9) and are among Adams' most
known and reprinted works (Fig. 1, and for histograms please
refer to appendix). Sample size was 25 cmx 30 cm (the same
size as the printsin the book), with a 3cm wide neutral,18%
reflectance grey masking. The prints were laminated, with a
neutral, transparent lamination, to avoid strong reflections from
the print surface and to protect the samples from wearing out-
conditions, which are liable to influence participants' judgment.
2.3 Stimuli Reproduction Process
The photographs were scanned in an "Epson" scanner GT-
9700F. For each photograph, a sample set composed of two
prints with original tones and three sets of 10 prints, for each of
the three curves was composed, as shown in Fig. 2:
"OR" is the central straight line corresponding to the repro-
duction of the original,
(1) A curve for "SH" region (toe), where contrast was
increased in the shadows and compressed the highlights.
As a result the visual impression is that the images look
lighter.
(2) A curve for "HI" region (shoulder), where contrast was
increased in highlights and compressed the shadows. The
resulting visual impression is the overall darkening of the
images.
(3) A curve for "MT" region (straight line), where contrast was
increased in the mid-tones, while compressed both high-
lights and shadows.
For all three regions ("HI", "SH" and "MT") contrast
increment ranged between 1% and 10% in increments of 1%
(see Fig. 2), in Photoshop curve function, named 1 to 10
accordingly.
The curveswere appliedto the scannedgrey scaleand photo-
graphsusing Photoshopsoftware,and prints were producedby
Lambdasystem- in a silver gelatin process, on a photographic
black-and-whitepaper. Lambda prints are made on a Durst
Lambdaprinter,which uses three coloredlasers to exposetra-
ditionalphotographicmedia.These prints havethe advantageof
using the same rich RGBcolor space employed by computer
monitors.In addition,these prints are free of dots sinceunlike
inkjet printers, the laser outputs are continuouslymodulated
rather than switchedon and off.Their resolutionis comparable
to 1200dpiscreenedoutput.
For calibrationof the reproduction system, the densities of
the originalgrey-scaleand its reproductionwere measuredin a
densitometer and plotted against each other to yield a linear
relationship.
3. Subjects
Subjectsbelongedto twogroups
1. 18 observers who were skilledin image evaluationtasks,

4. S. GERSHONI and H. KOBAYASHI " How We Look at Photographs" 521
photo #1 photo #2 photo #3
Photo #4 photo #5 photo #6
Photo #7 photo #8 photo #9
Fig. 1 Photographs used as stimuli for the sorting task (Experiment 1)
Nine photographs by the photographer Ansel Adams, "Photographs of the Southwest", New-York Graphic Society, Boston Massachusetts that
were used as stimuli:
1. "Spanish Peaks, Colorado", 1951, page 18
2. "Canyon de Chelly,Nationa Monument, Arizona", 1947, page 30
3. "Moonrise, Hernandez, New-Mexico", 1944, page 55
4. "Navajo Woman, Wide Ruin, Arizona", 1948, page 39
5. "Maynard Dixon, painter, Tucson, Arizona", 1944, page 16
6. "Martha Porter, pioneer woman, Ordeville, Utah", 1961,page 81
7. "Adobe Dwellings, Northern New-Mexico", 1958,page 49
8. "Santuario de Chimayo, New-Mexico", 1950, page 50
9. "Arches, North Court, Mission San Xavier del Bac, Tucson Arizona, 1968, page 94
named: "trained" group.
2. 15 inexperienced observers, named: "novice" group.
Average age of subjects for "trained" and "novice" groups was
25 and 27.83% of the trained subjects and 37% of the novice
subjects were practicing photography. 50% of the trained sub-
jects and 10% of the novice were familiar with Ansel Adams
work and 28% of the trained and 5% of novice reported to have
previously seen the photographs used as stimuli.
4. Experimental Conditions
Lighting: D-50 (600 lux)
Grey-scale order-rank experiment was 30 minutes per ses-
sion: sorting time for each of the three greyscale sets was
5 minutes, and 2 intermissions of 5 minutes between sets. Pho-
tographs sorting experiment was 30 minutes per session, with-
out intermission- time for sorting each set was 10 minutes.
The task was in three sequencing sessions. Subjects were
trained to discriminate a subset of two samples: with original
tone and maximum contrast increment applied.

5. 522 J.Soc.Photogr.Sci. Technol.Japan Vol.68No.6 (2005)
Fig. 2a Contrast increment curves applied to SH and HI regions
Fig. 2b Contrast increment curves applied to MT region
5. Procedure
Experiment 1: Grey-Scales Rank Order Task 49)
14 trained subjects were individually presented with grey
scale sets for regions: "HI", "SH", "MT". Participants were
instructed to arrange the grey-scales in each set from the grey-
scale with the lowest contrast at the left end of the row to the
one with the highest contrast in right end (Prototype for the
"Rank Order Scaling" is modeled after Bartleson and F Gru
m
Eds, 1984) 48).
Experiments 2 and 3: Photographs Sorting Task 49)
For each photograph, a set composed of the sets for regions
"MT"
, "HI" and "SH" plus "OR" (altogether 3lsamples) were
randomly mixed and arranged in a pile, placed in the center of
the table. Another "OR" sample was set as a reference, to which
each of the samples in the pile was compared. On the right of the
pile was an area marked "harder" (which in Japanese means
"higher contrast") •for all samples
, judged to have higher
contrast than the reference. On the left of the pile was the
area marked "softer"("lower contrast" in Japanese") •for all
samples, judged to have less contrast than reference or equal to
it. To reduce observer criterion drift (observer internal defini-
tion) and create a typical observer condition, definition and
usage of the terms "softer" and "harder" in relation to the term
"contrast" were clarified before each experiment 49)
.
6. Results and Discussion
Experiment 1: Grey-Scales Rank Order Task
Subjects generally showed more correct rank order rates for
contrast increments applied to HI and MT region sets of grey-
scales than to SH, and mean correct response rates were 8.57,
1.21 and 8.5, respectively, as shown in Fig. 4. For all three
sets, correct ranking increased systematically as a function of
contrast increments (especially samples 7 to 10). Because indi-
vidual subjects exhibited different rates of correct rank order,

6. S.GERSHONI and H. KOBAYASHI " How We Look at Photographs" 523
Highlight-GraySCales Set:
Shadows-Grayscales Sets:
Midtone-Grayscales Scts:
Fig. 3 Grey-scale used as stimuli of the ordinal task (Experiments 2
and3)
relative responses were averaged across the subjects. This
yielded a comparison between the three region sets, revealing
that correct ranking rates for HI and MT were substantially
higher than SH, as shown in Fig. 5. In all results and figures
described 1%-10% contrast increments (see stimuli prepara-
tion) are indicated by sample numbers 1 to 10.
The correct ranking ratio for HI ranged between 0.64 and
0.71 for samples 5 and 6, and 1.00 for samples 8 to 10. Similarly,
for MT set, ratio ranged between 0.64 for sample 2, 0.86 for
samples 8 and 9, and 0.93 for 10. However, the ratio for SH set
was substantially lower, ranging between 0.07 for samples1 to
3 contrast increments, and 0.21 for samples 4, 6 and 8 to 10.
In order to confirm these observations, a one-way analysis of
variance (ANOVA), with region (HI vs. SH vs. MT) as a variable,
was conducted. In this and all other statistical tests, alpha level
of 0.05 was used. There was a significant main effect of region
[F(2,26)=58.56].Inadditionapairedcomparisonofcontrastsof
HI and MT with SH, revealed the differential control of the
region, to which contrast increments were applied, over dis-
criminationofcontrastincrements [F(2,26)=87.85].
Experiment 2: Photographs Sorting Task with Trained
Subjects
(a) Discrimination ratios for categories "portraits", "land-
scape" and "Architecture", separately calculated, are shown as
afunctionoftheregions"HI","SH","MT"(with1%-10%
contrast increments, indicated by sample numbers 1 to 10) in
the three graphs of Fig. 6.
We expected to find a differential discrimination ratio as an
effect of category, such as, perhaps, a higher response rate for
portraits than for the other categories (Papathomas and Bono,
2004; also: Dolan et al., 1997 ) 50)51),based also on the findings
that meaningfulness and knowledge of specific object shape
interfere with attention and visual organization processes (see
introduction: Peterson and Gibson, 1991, 1993, 1994) 25-30);
also that features such as junctions are not solely processed by
low level mechanism but necessitate the use of global processes
and are not detected prior scene interpretation (McDermott,
2004) 52).However, we observed a general similarity in contrast
discrimination performance for all categories examined. Further-
more, variations in discrimination ratios were more substantial
among region sets within the categories than between catego-
ries. This suggests a stronger control of configural properties
of the stimuli over contrast discrimination performance, than
meaningfulness of the category.
The category effect interaction is shown as a function of the
regions in Fig. 7. To explore the effects of category and region,
a two-way analysis of variance (ANOVA), with categories
(portrait vs. landscape vs. architecture) and regions of contrast
increment (HI vs. SH vs. MT) as variables, was conducted.
The effect of category was not significant [F(2, 34)=1.93],
reflecting that there is no differential control of such conceptual
content over discrimination of contrast increments in the exam-
ined regions. In addition, there were significant main effects of
region [F(2, 34)=96.22]. Paired comparisons of contrasts
revealed that while discrimination performance did not differ
much between SH and HI regions [F(1, 34)=2.45], response
ratio significantly increased at MT region [F(1, 34)=189.99].
(b) An unexpected finding was that discrimination ratio was
lower at SH region of "landscape" category compared with the
other two categories, shown in Fig. 7. In order to investigate this

7. 524 J.Soc.Photogr.Sci.Technol.Japan Vol.68No. 6(2005)
Fig. 4 Mean relative correct responses for rank-order task with region-sets (HI, MT, SH) of grey-scales (Experiment 1)
Fig. 5 Comparison between means correct rank-order for region-sets (HI, SH, MT) of grey-scales (Experiment 1)
difference we compared the discrimination ratios of landscape
stimuli #1 and #3; both stimuli are similar in composition,
depicting a wide view of a plane in the desert, with distant low
mountains, and proportion between the area of 'sky' and 'land'
is approximately 2:1 respectively. The main difference is that
stimuli #1 depicts a light view, while stimuli #3 depicts a night
view (see section 2.2). The "portrait" graph in Fig. 6 shows
similar discrimination ratios at HI and MT regions of #1 and #3.
On the other hand, discrimination ratio at SH decreased to zero.
It is especially notable, since it did not occur at any other stimuli.
According to a two-way analysis of variance (ANOVA), with
landscape stimuli (light #1 vs. night #3) and regions (HI vs. SH
vs. MT) as variables, there was a significant main effect of
region [F(2,34)=104.65]. In addition, the interaction was sig-
nificant [F(2,34)=5.17], as shown in Fig. 8. Paired comparisons
of contrasts revealed that discrimination performance was
better at SH region for stimulus #1 (light) than #3 (night)
[F(1,34)=11.35]. One way to explain this finding is that when an
image is overall dark (night scenes), the increment in contrast
applied to HI region (visually perceived as a further darkening of
the image) is conceived as an increase in contrast, while, the
visual effect of increment in contrast applied to SH region
(resulting in the impression of the image as becoming less dark),
is perceived as a decrease in contrast.
Experiment 3: Photographs Sorting Task with Novice
Subjects
(a) Discrimination ratios of for categories "portraits", "land-
scape" or "Architecture" separately calculated, are shown as a
function of the regions "HI", "ST", "MT" in the three graphs of
Fig. 9. Novice subjects' performance of contrast discrimination
resembled the trained subjects (see Fig. 6) in respect to
variation in discrimination ratios being more substantial among
samples within the categories than between them. The category
effect interaction plot is shown as a function of the region in

8. S. GERSHONI and H. KOBAYASHI " How We Look at Photographs" 525
A. "Portrait" Category
B. "Landscape" Category
C. "Architecture" Category
Fig. 6 Mean relative contrast increment discrimination for region sets (HI, SH, MT) ofphotographs (Experiment 2,trained subjects)

9. 526 J.Soc.Photogr.Sci. Tecknol.JapanVol.68No. 6(2005)
Fig. 7 Category effect interaction plot for region-sets of photographs in categories: portrait, landscape and architecture (Experiment 2, trained subjects).
Fig. 8 Light vs. night scenes effect interaction plot for stimuli #1 and #3(Experiment 2, trained subjects)
Fig. 10. To explore the effects of category and region, a two-way
analysis of variance (ANOVA),with categories (portrait vs. land-
scape vs. architecture) and regions of contrast increment (HI vs.
SH vs. MT) as variables was conducted. The effect of category
was not significant [F(2,28)=0.67], reflecting that there is no
differential control of conceptual category over discrimination
of contrast increments in any of the regions. In addition there
were significant main effects of region [F(2,28)=36.24] and
P<0.001. Paired comparisons of contrasts revealed that
response to contrast increments in the SH and HI regions differ
less [F(1,28)=11.93] and P=0.018, than the enhanced response
for MT region [F(1,28)=60.55].
These results are consistent with the previous observations
(for trained subjects) that there is a stronger control of con-
figuration attributes over contrast discrimination performance,
than of category, since, unlike trained subjects, who are skilled
in performing image evaluation tasks, and might have been
intentionally neglecting the conceptual content of the stimuli and
concentrating on configuration attributes, such as edges, etc,
novice subjects have no earlier experience in such tasks,
therefore their response suggests that the response rate is
independent of acquired skill.
(b) One of the substantial differences in performance between
the trained and novice groups of subjects is in the response ratio
for contrast increments applied to SH region in the "Landscape"
category day vs. night, stimuli #1 and #3 (see above results
6.2.b.), where, unlike the trained group, novice group's category
effect interaction plot revealed that there was no significant
difference between the two stimuli. In addition, while the trained
group showed a mean average response ratio for SH 9.13 and
for HI 6.93, the novice group response ratio for HI was 13.49
and for SH only 6.69.This suggests an effect of skill (training)
over the interpretation of contrast increments, when applied to
HI or SH regions. To confirm these observations a two-way

10. S.GERSHONI and H. KoBAYAsHI "How We Look at Photographs" 527
A. "Portrait"Category
B. "Landscape"Category
C. "ArchitectuTe"Category
Fig. 9 Mean relative contrast increment discrimination for region-sets (HI, SH, MT)of photographs (Experiment 3, novice subjects)

11. 528 J.Soc.Photogr.Sci.Technol.Japan Vol.68No. 6(2005)
Fig. 10 Category effect interaction plot for region-sets of photographs in categories: portrait, landscape and architecture (Experiment 3, novice subjects).
Fig. 11 Group effect interaction plot for region-sets HI and SH of photographs (Experiments 2 and 3, novice vs. trained)
analysis of variance (ANOVA),with groups (trained vs. novice)
and regions (HI vs. SH) as variables was conducted. The inter-
action was significant [F(1, 2)=1293.24], reflecting the differen-
tial control of training over the responding to contrast incre-
ments in regions HI and SH. Paired comparison of contrasts
revealed that trained subjects responded significantly more
to SH than to HI [F(1, 2)=154.55], and more enhanced was
the novice subjects differential response to HI than to SH
[F(1,2)=1476.53]. Furthermore, a pair comparison between the
groups revealed that novice subjects responded to HI signifi-
cantly more than did the trained subjects [F(1,2)=1375.54],
whereas trained responded to SH significantly more than novice
subjects did [F(1, 2)=189.59], as shown in Fig. 11. A suggested
explanation is that while novice subjects interpret the overall
darkening of an image (for contrast increments applied to HI
region) as an increase in contrast, trained subjects, in contrary,
interpret the overall lightening of an image as an increase in
contrast. Both groups, however, showed no significant difference
in response ratio for "MT" region.
(c) Another difference between the two groups of subjects is
the false alarms ratios- when discriminating OR from OR (see
section 2.3). The mean average for false alarms ratio of the
novice group was 0.18, whereas that of the trained group was
only 0.07, as shown in Figs. 6 and 9 (the unit at the extreme left
on x-axis refers to OR). This observation was confirmed by a
one-way analysis of variance (ANOVA) with group (novice vs.
trained) as a variable, which revealed that the rate of false
alarms differed between novice and trained groups [F(1,8)=
12.46], reflecting the effect of training over false alarms in
performance of contrast discrimination, and suggesting a better
performance training.

12. S.GERSHONI and H. KOBAYASHI "How We Look atPhotographs" 529
7. General Discussion
The present results can be summarized as follows. In Experi-
ment 1, subject discriminated contrast increments applied to
three regions (highlights, shadows and mid-tones) of separate
grey-scale sets. They generally showed higher correct response
rates to increments in highlights and mid-tones than shadows.
In Experiment 2 trained subject discriminated contrast incre-
ments applied to natural scene photographs of three categories
(portrait, landscape and architecture). Subjects showed a sys-
tematic increase in response as a function of contrast increments
at all regions of all stimuli examined. However, the discrimi-
nation ratio was significantly higher at MT than SH and HI
with no significant difference between the last two. Further-
more, in Experiment 3, experimentally naïve subjects performed
the same task as in Experiment 2, and generally showed the
same discrimination pattern as the trained subjects. Two main
differences between subject groups were found. Novice subjects'
false alarms rates were more frequent than the trained subjects.
And an even more interesting difference was that novice sub-
jects responded to HI more than the trained subjects, whereas
trained subjects responded to SH more than the novice subjects.
Nevertheless, the findings of Experiments 2 and 3 converge on
the conclusion that the differences in contrast discrimination at
HI vs. SH between subject groups was on the basis of inter-
pretation and intentional judgment (see 6.3.c), rather than skill,
obtained by training, as both groups showed priority of spatial
configuration properties over category. One can view this differ-
ential interpretation of contrast as yet another evidence for the
interference of higher, cognitive processes, such as interpreta-
tion and intention, in lightness perception, although some other
higher processes, such as grouping, are not prior to lightness
constancy, but later processes, composed of both low and
high processing (Beck, 1975; Olson and Atteneave, 1970) 22)23).
This view is consistent with the approach that although per-
ceptual processing proceeds from local to global wholes, people
may gain conscious access to the results in the opposite order
(Marcel, 1983)53),and affect them.
The main conclusion of the present study is discussed by
means of the comparison between contrast discrimination for
grey-scales (Experiment 1) and photographs (Experiment 2
and 3). This comparison revealed a difference mainly at SH
region. In grey-scales the high contrast discrimination ratios for
MT and HI region sets and the extremely low ratios for SH,
suggest a better ability to discriminate contrast increments in
the mid-tones (with compressed highlight and shadow tones), or
in highlights (with compressed shadow tones), where perceived
impression is of the darkening of the image, than in shadows
(with compressed highlights), where the impression is the
overall lightening of the image. However, in photographs, con-
trast discrimination ratios at SH region sets, were significantly
improved, that is, the contrast increment in shadows with
compressed highlights, did not result in the same low ratios as
in greyscales, despite the perceived lightening of the image.
This is in line with "Anchoring Theory of Lightness Perception"
(Gilchrist and Cataliotti, 1994; Gilchrist, A., and Bonato, F,
1995) 17)18)47).According to anchoring theory, the grey-scales
and photographs can be described in terms of global and local
frameworks, which weighing, as stronger or weaker anchors
depends on factors, such as grouping, configuration, luminance
gradient, articulation and insulation. Thus, it is possible to
explain the significant difference between greyscales and
photographs, in terms of spatial properties and the resulting
differences in the strength of their local frameworks as
anchors, rather than in terms of meaningfulness of conceptual
content. Local frameworks in grey-scales are much weaker than
in photographs, due to their line configuration, low articulation
(of only 20 patches), and the low luminance gradient-0.1 density
increments between adjacent patches. These properties suggest
that in grey-scales the compromise between the global and local
frameworks resulted in a weak anchoring to the local framework
than the global one and a luminance matching with poor light-
ness constancy. In photographs though, the high articulation,
complex configuration and the abundance of edges with various
luminance gradients, are factors that strengthen anchoring to
local frameworks and result in better lightness constancy.
Thus, the low ability to discriminate contrast increments,
applied to shadow region in grey-scales, may be explained as the
result of weak local frameworks together with enlarged "white"
area, which strengthen the global framework luminance assign-
ment. That may also explain the perceived impression of the
lightening of the image. However, in photographs, although the
same lightening impression occurred, the complex configuration
and insulation of the white patches and the increased local con-
trast, strengthened local frameworks and improved lightness
constancy. Hence, discrimination at shadows improved. This
view is consistent with recent results, showing that lightness
constancy increases when the contrast at contextual edge is
larger than the mediating or background one (Soranzo and
Agostiini, 2004) 54);see also introduction.
The present study is the first systematic investigation of
lightness perception in art photographs, which greater scope is
the investigation of the connection between visual processing
and the aesthetic experience of art. The main conclusion from
the above discussion is that contrast discrimination improves
as a function of better anchoring to local frameworks. One of
the main factors that strengthen local frameworks' anchoring is
the abundance of local contrasts. Ansel Adams, in his photo-
graphs, methodologically applied the zone system to created
highly articulated and tonally rich images and enhanced what is
"saw and felt" by controlled local-contrast . We further assume
that images, which offer better lightness constancy, amplify the
aesthetic experience, as evident by the intentional choice made
by the photographer. This is in line with recent approaches
about the neural basis of art and aesthetic experience1)2)55),
according to which the basis of aesthetic experience is biologi-
cally motivated and is a manifestation of the brain mechanisms
and constraints. Additional research, using tasks for preference
and evaluation of photographs with contrast changes, and find-
ing the relation between ability to detect contrast changes and
preference, is needed to examine this assumption.
Acknowledgements
The authors wish to express their gratitude to M., Tsukada,
Horiuchi Color, Tokyo for printing and laminating the entire
stimuli and supporting the research and M., Jitsumori, Depart-
ment of Cognitive and Information Sciences, Chiba University,
Japan, for her teaching, much assistance and advice, R.,
Yamamoto for her helpful suggestions.

13. 530 J.Soc.Photogr.Sci.Technol.Japan Vol.68No. 6(2005)
Appendix
Histograms of stimuliwith 10%(maximal)
contrastincrement inregions: HI, SH, MT
vs. the unaltered stimuli OR(from leftto
right accordingly):
Stimuli#1
Stimuli#2
Stimuli#3
Stimuli#4
StimuIi#5
Stimuli#6
Stimuli#7
Stimuli#8
Stimuli#9
* Please note that the Shadows are at the le負 of the histogram and highlights at the right
.

14. S. GERSHONI and H. KOBAYASHI "How We Look at Photographs" 531
References
1) S. Zeki, 'Art and the Brain', Oxford University Press, 2000, ISBN
0-19-850519-1.
2) V. S. Ramachandran, W. Hirstein, 'The Science of Art-A Neuro-
logical Theory of Aesthetic Experience', Journal of Consciousness
Studies, 6(6-7), 15-51 (1999).
3) R. McClelland, 'The Photographer and the Natural World', The
British Journal of Photography annual, 145-152 (1972).
4) The transferability thesis by Currie (1985), G. Currie, 'The authentic
and the aesthetic' American Philosophical Quarterly, 22, 153-160
(1985).
5) J. Elkins, 'What we are seeing exactly?', Art Bulletin, 79, 191-198
(1997).
6) H. Helson, V Joy, 'Domains of lightness, assimilation and contrast
effects in vision', Psychologische Beitraege, 6,405-415 (1962).
7) H. Helson, 'Adaptation-Level Theory', New-York, Harper & Row,
1964.
8) H. Wallach, 'Brightness constancy and the nature of achromatic
colors', Journal of Experimental Psychology, 38, 310-324 (1948).
9) H. Leibowitz, 'Visual Perception', New-York,MacMillan, 1965.
10) D. Jameson, L. M. Hurvich, 'Theory of brightness and color contrast
in human vision', Vision Research, 4, 135-154 (1964).
11) H. H. Land, J. J. McCann, 'Lightness and Retinex theory', Journal of
Optical Society of America, 61, 1-11 (1971).
12) D. Knill, D. Kersten, 'Apparent surface curvature affects lightness
perception', Nature, 351, 228-230 (1991).
13) A. L. Gilchrist, 'Perceived lightness depends on perceived spatial
arrangement', Science, 195(4274), 185-187 (1977).
14) H. G. Barrow,J. Tenenbaum, 'Recovering intrinsic scene characteris-
tics from images', in 'Computer Vision System' (A. R. Hanson & E.
M. Riseman), Orlando, Academic Press, 3-26 (1978).
15) L.Arend, 'Surface colors, Illumination, and surface geometry: Intrin-
sic Image Models of Human Color Perception, in 'Lightness, Bright-
ness and Transparency', (A. Gilchrist), Hillsdale, Erlbaum, 159-213
(1994).
16) E. H. Adelson, A. P. Pentland, 'The perception of shading and re-
flectance', in 'Perception as Bayesian Inference', (D. Knill and W.
Richards), New-York, Cambridge University Press, 409-423 (1996).
17) A. L. Gilchrist, J. Cataliotti, 'Anchoring of surface lightness and
multiple illumination levels', Investigative Opthalmology and Vision
Science, 35(4), 2165 (1994).
18) A. Gilchrist, E Bonato, 'Anchoring of lightness values in center/
surround displays', Journal of Experimental Psychology, 21(6),
1427-1440 (1995).
19) R. 0. Brown, 'The World is not Grey', Investigative Opthalmology
and Vision Science, 35(4), 2165 (1994).
20) J. McCann, 'Psychological Experiments in Search of Adaptation and
the Gray World', IS & T Annual Meeting, Rochester, New York,
1994.
21) J. A. Schirillo, S. Shevel, 'Brightness contrast from inhomogeneous
surrounds', Vision Research, 36(12), 1783-1796 (1996).
22) J. Beck, 'The relation Between Similarity Grouping and perceptual
constancy', American Journal of Psychology, 88(3), 397-409 (1975).
23) R. K. Olson, E Attneave, 'What variables produce similarity group-
ing?', American Journal of Psychology, 83(1), 1-21 (1970).
24) I. Rock, R. Nijhawan, S. Palmer, L. Tudor, 'Grouping Based on
Phenomenal similarity of achromatic color, Perception, 21(6), 779-
789 (1992).
25) S. E. Palmer, J. Neff, D. Beck, 'Late influences on perceptual group-
ing', Psychonomic Bulletin & Review, 3(1), 75-80 (1996).
26) J. Shi, J. Mlik, Normalized cuts and image segmentation, IEEE
Conference on Computation: Vision and Pattern Recognition, San
Juan, Puerto Rico, in, 'Vision Science', (Stephen, E. Palmer),
Cambridge Massachusetts, MIT Press, 1999,271-272 (1997).
27) M. Wetheimer, 'Gestalt Theory', in, 'Vision Science', (Stephen, E.
Palmer), Cambridge Massachusetts, MIT Press, 1999,256-261,263,
266-267,472-473 (1924).
28) M. A. Peterson, B. S. Gibson, 'The initial identification of figure/
ground relationships: Contributions from shape recognition process',
Bulletin of Psychonomic Society, 29(3), 199-202 (1991).
29) M. A. Peterson, B. S. Gibson, Shape recognition inputs to figure-
ground organization in three dimensional grounds, Cognitive
Psychology,25(3), 383-429 (1993).
30) M. A. Peterson, The proper treatment of uniform connectedness',
Psychological Bulletin & Review, 1, 509-514 (1993).
31) I. Rock, The Logic of Perception', in 'Vision Science', (Stephen, E.
Palmer), 1983, Cambridge Massachusetts, MIT Press, 1999, 264-
265,268-270,274-275,281,284-286,299.
32) D. G. Lowe, 'Perceptual Organization and Visual Recognition, 1985,
in, 'Vision Science', (Stephen, E. Palmer), Cambridge Massachu-
setts, MIT Press, 1999,299.
33) M. K. Albert, D. D. Hoffman, `Genericity in spatial vision', 1995, in
'Vision Science', (Stephen, E. Palmer), Cambridge Massachusetts,
MIT Press, 1999,293,299.
34) H. Peterhans, R. von der Heydt, 'Mechanisms of contour perception
in monkey visual cortex. II. Contours Bridging Gaps', Trend in
Neuroscience, 9(5), 1749-1763 (1989).
35) H. Peterhans, R. von der Heydt, 'Subjective contours: Bridging the
gap between psychophysics and physiology', Journal of Neuro-
science, 14(3), 112-119 (1991).
36) S. P Vecera, R. C. O'Reilly, 'Figure-ground organization and object
recognition processes', Journal of Experimental Psychology: Human
Perception & Performance, 24(2), 441-462 (1998).
37) H. Mitsumatsu, K. Yokosawa, 'How do the internal details of the
object contribute to recognition?', Perception, 31(11), 1289-1298
(2002).
38) D. Marr, E. C. Hildreth, 'Theory of edge detection', in, 'Vision
Science', (Stephen, E. Palmer), Cambridge Massachusetts, MIT
Press, 270,277 (1980).
39) S. E. Palmer, I. Rock, 'On the nature and order of organizational
processing', Psychonomic Bulletin & Review, 1, 515-519 (1994).
40) S. E. Palmer, I. Rock, 'Rethinking perceptual organization: The role
of uniform connectedness', Psychonomic Bulletin & Review, 1(1),
29-55 (1994).
41) L. J. Thompson, E. P Chronicle, A. E Collins, 'The role of pictorial
convention in haptic picture perception', Perception, 32(7), 887-893
(2003).
42) A. L. Gilchrist, 'Perceived Lightness Depends on Perceived Spatial
Arrangement', Science, 195(4274), 185-187 (0000).
43) A. L. Gilchrist, 'When Does Perceived Lightness Depend on
Perceived Spatial Arrangement?', Perception and Psychophysics,
28(6), 527-538 (1977).
44) A. Adams, 'Photographs of the Southwest', NewYork Graphic
Society, Boston Massachusetts, 1976.
45) A. Hammond, 'Ansel Adams and the Objectivism-Making a Photo-
graph With Group f/64', History of Photography, 22(2), 169-178
(1998).
46) H. L. Ira, 'Ansel Adams, The Zone System and the California School
of Fine Arts', History of Photography, 22(2), 147-154 (1988).
47) A. Gilchrist, C. Kossyfidis, 'An Anchoring Theory of Lightness
Perception', Psychological Review, 106(4), 795-834 (1999).
48) C. J. Bartleson, Grum F Eds., 'Optical Radiation Measurements',
Orlando: Academic Press, 1984, Vol.5, Chapter 8, in 'Psychometric
Scaling: A Toolkit For Imaging Systems Development', (Peter
Engeldrum), Winchester, MA: Imcotec Press, 79-83 (2000).
49) P Engeldrum, Psychometric Scaling: A Toolkit For Imaging Systems
Development", Winchester, MA: Imcotec Press, 63-86 (2000).
50) P.V Papathomas, L. M. Bono, 'Experiments with a hollow mask and
a reverspective: Topdown influences in the inversion effect for 3-D
stimuli', Perception, 33(9), 1129-1138 (2004).
51) R. J. Dolan, et al. , 'How the brain learns to see objects and faces in an
impoverished context, Nature, 596-599 (1997).
52) J. McDermott, Psychophysics with junctions in real images',
Perception, 33(9), 1101-1127 (2004).
53) A. J. Marcel, 'Conscious and unconscious perceptions: An approach
to the relations between phenomenal experience and perceptual
processes', Cognitive Psychology, 15, 197-300 (1983).
54) A. Soranzo, T.Ggostini, 'Impossible shadows and lightness constancy',
Perception, 33(11), 1359-1368 (2004).
55) S. Zeki, 'Neural Concept Formation & Art', Journal of Consciousness
Studies, 9(3), 53-76 (2002).