This document proposes a new image representation called "blobworld" for content-based image retrieval. It uses EM segmentation on combined color and texture features to segment images into coherent "blobs". The system allows users to view an image's internal blobworld representation to better understand query results. It aims to improve on existing systems by recognizing images as combinations of objects rather than just "stuff".
Content-based Image Retrieval Using The knowledge of Color, Texture in Binary...Zahra Mansoori
This document presents a new approach for content-based image retrieval that combines color, texture, and a binary tree structure to describe images and their features. Color histograms in HSV color space and wavelet texture features are extracted as low-level features. A binary tree partitions each image into regions based on color and represents higher-level spatial relationships. The performance of the proposed system is evaluated on a subset of the COREL image database and compared to the SIMPLIcity image retrieval system. Experimental results show the proposed system has better retrieval performance than SIMPLIcity in some categories and comparable performance in others.
An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance FeedbackIJMIT JOURNAL
Content-based image retrieval (CBIR) systems utilize low level query image feature as identifying similarity between a query image and the image database. Image contents are plays significant role for image retrieval. There are three fundamental bases for content-based image retrieval, i.e. visual feature extraction, multidimensional indexing, and retrieval system design. Each image has three contents such as: color, texture and shape features. Color and texture both plays important image visual features used in Content-Based Image Retrieval to improve results. Color histogram and texture features have potential to retrieve similar images on the basis of their properties. As the feature extracted from a query is low level, it is extremely difficult for user to provide an appropriate example in based query. To overcome these problems and reach higher accuracy in CBIR system, providing user with relevance feedback is famous for provide promising solutio
This document summarizes an evaluation of texture feature extraction methods for content-based image retrieval, including co-occurrence matrices, Tamura features, and Gabor filters. The evaluation tested these methods on a Corel image collection using Manhattan distance as the similarity measure. Co-occurrence matrices performed best with homogeneity as the feature, while Gabor wavelets showed better performance for homogeneous textures of fixed sizes. Tamura features performed poorly with directionality. Overall, co-occurrence matrices provided the best results for general texture retrieval.
This document proposes a new method for segmenting outdoor images called Color Cluster Elimination (CCE) which utilizes color clustering and texture analysis. CCE performs color clustering in a multi-resolution pyramid to gradually eliminate larger color clusters, preventing them from dominating segmentation and allowing smaller clusters to emerge more clearly. It then examines regions for adjacent homochromatic objects with different textures, introducing Texture Sewn Response (TSR) to indicate texture strength across resolutions/directions. The method is evaluated on the BSDS500 dataset against other metrics, demonstrating satisfactory performance for outdoor scene segmentation.
This document describes a proposed content-based image retrieval system using backpropagation neural networks (BPNN) and k-means clustering. It begins by discussing CBIR techniques and features like color, texture, and shape. It then outlines the proposed system which includes training a BPNN on image features, validating images, and testing by querying and retrieving similar images. Performance is analyzed based on metrics like accuracy, efficiency, and classification rate. Results show the system achieves up to 98% classification accuracy within 5-6 seconds.
This document presents a method for interactive image segmentation using constrained active contours. It begins with an overview of existing interactive segmentation techniques, including boundary-based methods like active contours/snakes and region-based methods like random walks and graph cuts. The proposed method initializes a contour using region-based segmentation then refines it using a convex active contour model that incorporates both regional information from seed pixels and boundary smoothness. This allows the contour to globally evolve to object boundaries while handling topology changes.
The development of multimedia system technology in Content based Image Retrieval (CBIR) System is
one in every of the outstanding area to retrieve the images from an oversized collection of database. The feature
vectors of the query image are compared with feature vectors of the database images to get matching images.It is
much observed that anyone algorithm isn't beneficial in extracting all differing kinds of natural images. Thus an
intensive analysis of certain color, texture and shape extraction techniques are allotted to spot an efficient CBIR
technique that suits for a selected sort of images. The Extraction of an image includes feature description and
feature extraction. During this paper, we tend to projected Color Layout Descriptor (CLD), grey Level Co-
Occurrences Matrix (GLCM), Marker-Controlled Watershed Segmentation feature extraction technique that
extract the matching image based on the similarity of Color, Texture and shape within the database. For
performance analysis, the image retrieval timing results of the projected technique is calculated and compared
with every of the individual feature.
Comparative Study and Analysis of Image Inpainting TechniquesIOSR Journals
Abstract: Image inpainting is a technique to fill missing region or reconstruct damage area from an image.It
removes an undesirable object from an image in visually plausible way.For filling the part of image, it use
information from the neighboring area. In this dissertation work, we present a Examplar based method for
filling in the missing information in an image, which takes structure synthesis and texture sysnthesis together.
In exemplar based approach it used local information from an image to patch propagation.We have also
implement Nonlocal Mean approach for exemplar based image inpainting.In Nonlocal mean approach it find
multiple samples of best exemplar patches for patch propagation and weight their contribution according to
their similarity to the neighborhood under evaluation. We have further extended this algorithm by considering
collaborative filtering method to synthesize and propagate with multiple samples of best exemplar patches. We
have to preformed experiment on many images and found that our algorithm successfully inpaint the target
region.We have tested the accuracy of our algorithm by finding parameter like PSNR and compared PSNR
value for all three different approaches.
Keywords: Texture Synthesis, Structure Synthesis, Patch Propagation ,imageinpainting ,nonlocal approach,
collabrative filtering.
Content-based Image Retrieval Using The knowledge of Color, Texture in Binary...Zahra Mansoori
This document presents a new approach for content-based image retrieval that combines color, texture, and a binary tree structure to describe images and their features. Color histograms in HSV color space and wavelet texture features are extracted as low-level features. A binary tree partitions each image into regions based on color and represents higher-level spatial relationships. The performance of the proposed system is evaluated on a subset of the COREL image database and compared to the SIMPLIcity image retrieval system. Experimental results show the proposed system has better retrieval performance than SIMPLIcity in some categories and comparable performance in others.
An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance FeedbackIJMIT JOURNAL
Content-based image retrieval (CBIR) systems utilize low level query image feature as identifying similarity between a query image and the image database. Image contents are plays significant role for image retrieval. There are three fundamental bases for content-based image retrieval, i.e. visual feature extraction, multidimensional indexing, and retrieval system design. Each image has three contents such as: color, texture and shape features. Color and texture both plays important image visual features used in Content-Based Image Retrieval to improve results. Color histogram and texture features have potential to retrieve similar images on the basis of their properties. As the feature extracted from a query is low level, it is extremely difficult for user to provide an appropriate example in based query. To overcome these problems and reach higher accuracy in CBIR system, providing user with relevance feedback is famous for provide promising solutio
This document summarizes an evaluation of texture feature extraction methods for content-based image retrieval, including co-occurrence matrices, Tamura features, and Gabor filters. The evaluation tested these methods on a Corel image collection using Manhattan distance as the similarity measure. Co-occurrence matrices performed best with homogeneity as the feature, while Gabor wavelets showed better performance for homogeneous textures of fixed sizes. Tamura features performed poorly with directionality. Overall, co-occurrence matrices provided the best results for general texture retrieval.
This document proposes a new method for segmenting outdoor images called Color Cluster Elimination (CCE) which utilizes color clustering and texture analysis. CCE performs color clustering in a multi-resolution pyramid to gradually eliminate larger color clusters, preventing them from dominating segmentation and allowing smaller clusters to emerge more clearly. It then examines regions for adjacent homochromatic objects with different textures, introducing Texture Sewn Response (TSR) to indicate texture strength across resolutions/directions. The method is evaluated on the BSDS500 dataset against other metrics, demonstrating satisfactory performance for outdoor scene segmentation.
This document describes a proposed content-based image retrieval system using backpropagation neural networks (BPNN) and k-means clustering. It begins by discussing CBIR techniques and features like color, texture, and shape. It then outlines the proposed system which includes training a BPNN on image features, validating images, and testing by querying and retrieving similar images. Performance is analyzed based on metrics like accuracy, efficiency, and classification rate. Results show the system achieves up to 98% classification accuracy within 5-6 seconds.
This document presents a method for interactive image segmentation using constrained active contours. It begins with an overview of existing interactive segmentation techniques, including boundary-based methods like active contours/snakes and region-based methods like random walks and graph cuts. The proposed method initializes a contour using region-based segmentation then refines it using a convex active contour model that incorporates both regional information from seed pixels and boundary smoothness. This allows the contour to globally evolve to object boundaries while handling topology changes.
The development of multimedia system technology in Content based Image Retrieval (CBIR) System is
one in every of the outstanding area to retrieve the images from an oversized collection of database. The feature
vectors of the query image are compared with feature vectors of the database images to get matching images.It is
much observed that anyone algorithm isn't beneficial in extracting all differing kinds of natural images. Thus an
intensive analysis of certain color, texture and shape extraction techniques are allotted to spot an efficient CBIR
technique that suits for a selected sort of images. The Extraction of an image includes feature description and
feature extraction. During this paper, we tend to projected Color Layout Descriptor (CLD), grey Level Co-
Occurrences Matrix (GLCM), Marker-Controlled Watershed Segmentation feature extraction technique that
extract the matching image based on the similarity of Color, Texture and shape within the database. For
performance analysis, the image retrieval timing results of the projected technique is calculated and compared
with every of the individual feature.
Comparative Study and Analysis of Image Inpainting TechniquesIOSR Journals
Abstract: Image inpainting is a technique to fill missing region or reconstruct damage area from an image.It
removes an undesirable object from an image in visually plausible way.For filling the part of image, it use
information from the neighboring area. In this dissertation work, we present a Examplar based method for
filling in the missing information in an image, which takes structure synthesis and texture sysnthesis together.
In exemplar based approach it used local information from an image to patch propagation.We have also
implement Nonlocal Mean approach for exemplar based image inpainting.In Nonlocal mean approach it find
multiple samples of best exemplar patches for patch propagation and weight their contribution according to
their similarity to the neighborhood under evaluation. We have further extended this algorithm by considering
collaborative filtering method to synthesize and propagate with multiple samples of best exemplar patches. We
have to preformed experiment on many images and found that our algorithm successfully inpaint the target
region.We have tested the accuracy of our algorithm by finding parameter like PSNR and compared PSNR
value for all three different approaches.
Keywords: Texture Synthesis, Structure Synthesis, Patch Propagation ,imageinpainting ,nonlocal approach,
collabrative filtering.
Web Image Retrieval Using Visual Dictionaryijwscjournal
In this research, we have proposed semantic based image retrieval system to retrieve set of relevant images for the given query image from the Web. We have used global color space model and Dense SIFT feature extraction technique to generate visual dictionary using proposed quantization algorithm. The images are transformed into set of features. These features are used as inputs in our proposed Quantization algorithm for generating the code word to form visual dictionary. These codewords are used to represent images semantically to form visual labels using Bag-of-Features (BoF). The Histogram intersection method is used to measure the distance between input image and the set of images in the image database to retrieve similar images. The experimental results are evaluated over a collection of 1000 generic Web images to demonstrate the effectiveness of the proposed system.
Research Inventy : International Journal of Engineering and Scienceresearchinventy
Research Inventy : International Journal of Engineering and Science is published by the group of young academic and industrial researchers with 12 Issues per year. It is an online as well as print version open access journal that provides rapid publication (monthly) of articles in all areas of the subject such as: civil, mechanical, chemical, electronic and computer engineering as well as production and information technology. The Journal welcomes the submission of manuscripts that meet the general criteria of significance and scientific excellence. Papers will be published by rapid process within 20 days after acceptance and peer review process takes only 7 days. All articles published in Research Inventy will be peer-reviewed.
Content Based Image Retrieval : Classification Using Neural Networksijma
In a content-based image retrieval system (CBIR), the main issue is to extract the image features that
effectively represent the image contents in a database. Such an extraction requires a detailed evaluation of
retrieval performance of image features. This paper presents a review of fundamental aspects of content
based image retrieval including feature extraction of color and texture features. Commonly used color
features including color moments, color histogram and color correlogram and Gabor texture are
compared. The paper reviews the increase in efficiency of image retrieval when the color and texture
features are combined. The similarity measures based on which matches are made and images are
retrieved are also discussed. For effective indexing and fast searching of images based on visual features,
neural network based pattern learning can be used to achieve effective classification.
COLOUR BASED IMAGE SEGMENTATION USING HYBRID KMEANS WITH WATERSHED SEGMENTATIONIAEME Publication
Image processing, arbitrarily manipulating an image to achieve an aesthetic standard or to support a preferred reality. The objective of segmentation is partitioning an image into distinct regions containing each pixels with similar attributes. Image segmentation can be done using thresholding, color space segmentation, k-means clustering.
Segmentation is the low-level operation concerned with partitioning images by determining disjoint and homogeneous regions or, equivalently, by finding edges or boundaries. The homogeneous regions, or the edges, are supposed to correspond, actual objects, or parts of them, within the images. Thus, in a large number of applications in image processing and computer vision, segmentation plays a fundamental role as the first step before applying to images higher-level operations such as recognition, semantic interpretation, and representation. Until very recently, attention has been focused on segmentation of gray-level images since these have been the only kind of visual information that acquisition devices were able to take the computer resources to handle. Nowadays, color image has definitely displaced monochromatic information and computation power is no longer a limitation in processing large volumes of data. In this paper proposed hybrid k-means with watershed segmentation algorithm is used segment the images. Filtering techniques is used as noise filtration method to improve the results and PSNR, MSE performance parameters has been calculated and shows the level of accuracy
An Automatic Color Feature Vector Classification Based on Clustering MethodRSIS International
In computer vision application, visual features such as
shape, color and texture are extracted to characterize images.
Each of the features is represented using one or more feature
descriptors. One of the important requirements in image
retrieval, indexing, classification, clustering, etc. is extracting
efficient features from images. The color feature is one of the
most widely used visual features. Use of color histogram is the
most common way for representing color feature. One of
disadvantage of the color histogram is that it does not take the
color spatial distribution into consideration. In this paper an
automatic color feature vector classification based on clustering
approach is presented, which effectively describes the spatial
information of color features. The image retrieval results are
compare to improved color feature vector show the acceptable
efficiency of this approach. It propose an automatic color feature
vector classification of satellite images using clustering approach.
The intention is to study cluster a set of satellite images in several
categories on the color similarity basis. The images are processed
using LAB color space in the feature extraction stage. The
resulted color-based feature vectors are clustered using an
automatic unsupervised classification algorithm. Some
experiments based on the proposed recognition technique have
also been performed. More research, however, is needed to
identify and reduce uncertainties in the image processing chain
to improve classification accuracy. The mathematical training
and prediction analysis of a general familiarity with satellite
classifications meet typical map accuracy standards.
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...CSCJournals
The document describes an image segmentation algorithm that uses both color and depth features extracted from RGBD images captured by a Kinect sensor. The algorithm clusters pixels into segments based on their color, texture, 3D spatial coordinates, surface normals, and the output of a graph-based segmentation algorithm. Depth features help resolve illumination issues and occlusion that cannot be handled by color-only methods. The algorithm was tested on commercial building images and showed potential for real-time applications.
Engineering Research Publication
Best International Journals, High Impact Journals,
International Journal of Engineering & Technical Research
ISSN : 2321-0869 (O) 2454-4698 (P)
www.erpublication.org
Content Based Image Retrieval Using Dominant Color and Texture FeaturesIJMTST Journal
The purpose of this Paper is to describe our research on different feature extraction and matching techniques in designing a Content Based Image Retrieval (CBIR) system. Due to the enormous increase in image database sizes, as well as its vast deployment in various applications, the need for CBIR development arose. Content Based Image Retrieval (CBIR) is the retrieval of images based on features such as color and texture. Image retrieval using color feature cannot provide good solution for accuracy and efficiency. The most important features are Color and texture. In this paper technique used for retrieving the images based on their content namely dominant color, texture and combination of both color and texture. The technique verifies the superiority of image retrieval using multi feature than the single feature.
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
A Combined Method with automatic parameter optimization for Multi-class Image...AM Publications
Multi-class image semantic segmentation deals with many applications in consumer electronics
fields such as image editing and image retrieval. Segmentation is done by combining the top down and bottomup
segmentation. Top-Down Process can be done by Semantic Texton Forest and bottom up- process using
JSEG. These two segmentation process can be executed in a combined manner. But this cannot choose the
optimal value of JSEG parameter for each interested semantic category. Hence an automatic parameter selection
algorithm has been proposed. An automatic parameter selection technique called an automatic multilevel
thresholding algorithm using stratified sampling and PSO is used to remedy the limitations.
Image Segmentation Using Pairwise Correlation ClusteringIJERA Editor
A pairwise hypergraph based image segmentation framework is formulated in a supervised manner for various images. The image segmentation is to infer the edge label over the pairwise hypergraph by maximizing the normalized cuts. Correlation clustering which is a graph partitioning algorithm, was shown to be effective in a number of applications such as identification, clustering of documents and image segmentation.The partitioning result is derived from a algorithm to partition a pairwise graph into disjoint groups of coherent nodes. In the pairwise correlation clustering, the pairwise graph which is used in the correlation clustering is generalized to a superpixel graph where a node corresponds to a superpixel and a link between adjacent superpixels corresponds to an edge. This pairwise correlation clustering also considers the feature vector which extracts several visual cues from a superpixel, including brightness, color, texture, and shape. Significant progress in clustering has been achieved by algorithms that are based on pairwise affinities between the datasets. The experimental results are shown by calculating the typical cut and inference in an undirected graphical model and datasets.
This document presents a new framework for color image segmentation using a combination of watershed and seed region growing algorithms. It begins with an introduction to image segmentation and discusses challenges with traditional gray-scale methods when applied to color images. The document then proposes a method using automatic seed region growing integrated with the watershed algorithm. Experimental results on an input image are shown to demonstrate the segmentation process and output images. The framework is concluded to improve upon traditional gray-scale methods for segmenting the richer information in color images.
Automatic dominant region segmentation for natural imagescsandit
Image Segmentation segments an image into different homogenous regions. An efficient
semantic based image retrieval system divides the image into different regions separated by
color or texture sometimes even both. Features are extracted from the segmented regions and
are annotated automatically. Relevant images are retrieved from the database based on the
keywords of the segmented region In this paper, automatic image segmentation is proposed to
obtained dominant region of the input natural images. Dominant region are segmented and
results are obtained . Results are also recorded in comparison to JSEG algorithm
The project aims at development of efficient segmentation method for the CBIR system. Mean-shift segmentation generates a list of potential objects which are meaningful and then these objects are clustered according to a predefined similarity measure. The method was tested on benchmark data and F-Score of .30 was achieved.
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposureiosrjce
This document discusses boundary detection techniques for images. It proposes a generalized boundary detection method (Gb) that combines low-level and mid-level image representations in a single eigenvalue problem to detect boundaries. Gb achieves state-of-the-art results at low computational cost. Soft segmentation and contour grouping methods are also introduced to further improve boundary detection accuracy with minimal extra computation. The document presents outputs of Gb on sample images and concludes that Gb effectively detects boundaries in a principled manner by jointly resolving constraints from multiple image interpretation layers in closed form.
In Content-Based Image Retrieval (CBIR) systems, the visual contents of the
images in the database are took out and represented by multi-dimensional characteristic
vectors. A well known CBIR system that retrieves images by unsupervised method known
as cluster based image retrieval system. For enhancing the performance and retrieval rate
of CBIR system, we fuse the visual contents of an image. Recently, we developed two
cluster-based CBIR systems by fusing the scores of two visual contents of an image. In this
paper, we analyzed the performance of the two recommended CBIR systems at different
levels of precision using images of varying sizes and resolutions. We also compared the
performance of the recommended systems with that of the other two existing CBIR systems
namely UFM and CLUE. Experimentally, we find that the recommended systems
outperform the other two existing systems and one recommended system also comparatively
performed better in every resolution of image.
Literature Review on Content Based Image RetrievalUpekha Vandebona
This document summarizes a literature review on content-based image retrieval (CBIR). It discusses how CBIR uses computer vision techniques to automatically extract visual features from images for retrieval, unlike traditional concept-based methods that rely on metadata/text. The key visual features discussed are color, texture, and shape. A typical CBIR system architecture includes creating an image database, automatically extracting features, searching by example or semantics, and ranking results. Distance measures are used to compare image features and evaluate retrieval performance. Combining CBIR with concept-based techniques could improve image retrieval overall.
Semi-Supervised Method of Multiple Object Segmentation with a Region Labeling...sipij
Efficient and efficient multiple object segmentation is an important task in computer vision and object recognition. In this work; we address a method to effectively discover a user’s concept when multiple objects of interest are involved in content based image retrieval. The proposed method incorporate a framework for multiple object retrieval using semi-supervised method of similar region merging and flood fill which models the spatial and appearance relations among image pixels. To improve the effectiveness of similarity based region merging we propose a new similarity based object retrieval. The users only need to roughly indicate the after which steps desired objects contour is obtained during the automatic merging of similar regions. A novel similarity based region merging mechanism is proposed to guide the merging process with the help of mean shift technique and objects detection using region labeling and flood fill. A region R is merged with its adjacent regions Q if Q has highest similarity with Q (using Bhattacharyya descriptor) among all Q’s adjacent regions. The proposed method automatically merges the regions that are initially segmented through mean shift technique, and then effectively extracts the object contour by merging all similar regions. Extensive experiments are performed on 12 object classes (224 images total) show promising results.
WEB IMAGE RETRIEVAL USING CLUSTERING APPROACHEScscpconf
Image retrieval system is an active area to propose a new approach to retrieve images from the
large image database. In this concerned, we proposed an algorithm to represent images using
divisive based and partitioned based clustering approaches. The HSV color component and Haar wavelet transform is used to extract image features. These features are taken to segment an image to obtain objects. For segmenting an image, we used modified k-means clustering algorithm to group similar pixel together into K groups with cluster centers. To modify Kmeans, we proposed a divisive based clustering algorithm to determine the number of cluster and get back with number of cluster to k-means to obtain significant object groups. In addition, we also discussed the similarity distance measure using threshold value and object uniqueness to quantify the results.
Color and texture based image retrievaleSAT Journals
Abstract Content-based image retrieval (CBIR) is an vital research area for manipulating bulky image databases and records. Alongside the conventional method where the images are searched on the basis of words, CBIR system uses visual contents to retrieve the images. In content based image retrieval systems texture and color features have been the primal descriptors. We use HSV color information and mean of the image as texture information. The performance of proposed scheme is calculated on the basis of precision, recall and accuracy. As an effect, the blend of color and texture features of the image provides strong feature set for image retrieval. Keywords: image retrieval, HSV color space, color histogram, image texture.
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Ijaems apr-2016-16 Active Learning Method for Interactive Image RetrievalINFOGAIN PUBLICATION
With many possible multimedia applications, content-based image retrieval (CBIR) has recently gained more interest for image management and web search. CBIR is a technique that utilizes the visual content of an image, to search for similar images in large-scale image databases, according to a user’s concern. In image retrieval algorithms, retrieval is according to feature similarities with respect to the query, ignoring the similarities among images in database. To use the feature similarities information, this paper presents the k-means clustering algorithm to image retrieval system. This clustering algorithm optimizes the relevance results by firstly clustering the similar images in the database. In this paper, we are also implementing wavelet transform which demonstrates significant rough and precise filtering. We also apply the Euclidean distance metric and input a query image based on similarity features of which we can retrieve the output images. The results show that the proposed approach can greatly improve the efficiency and performances of image retrieval.
Web Image Retrieval Using Visual Dictionaryijwscjournal
In this research, we have proposed semantic based image retrieval system to retrieve set of relevant images for the given query image from the Web. We have used global color space model and Dense SIFT feature extraction technique to generate visual dictionary using proposed quantization algorithm. The images are transformed into set of features. These features are used as inputs in our proposed Quantization algorithm for generating the code word to form visual dictionary. These codewords are used to represent images semantically to form visual labels using Bag-of-Features (BoF). The Histogram intersection method is used to measure the distance between input image and the set of images in the image database to retrieve similar images. The experimental results are evaluated over a collection of 1000 generic Web images to demonstrate the effectiveness of the proposed system.
Research Inventy : International Journal of Engineering and Scienceresearchinventy
Research Inventy : International Journal of Engineering and Science is published by the group of young academic and industrial researchers with 12 Issues per year. It is an online as well as print version open access journal that provides rapid publication (monthly) of articles in all areas of the subject such as: civil, mechanical, chemical, electronic and computer engineering as well as production and information technology. The Journal welcomes the submission of manuscripts that meet the general criteria of significance and scientific excellence. Papers will be published by rapid process within 20 days after acceptance and peer review process takes only 7 days. All articles published in Research Inventy will be peer-reviewed.
Content Based Image Retrieval : Classification Using Neural Networksijma
In a content-based image retrieval system (CBIR), the main issue is to extract the image features that
effectively represent the image contents in a database. Such an extraction requires a detailed evaluation of
retrieval performance of image features. This paper presents a review of fundamental aspects of content
based image retrieval including feature extraction of color and texture features. Commonly used color
features including color moments, color histogram and color correlogram and Gabor texture are
compared. The paper reviews the increase in efficiency of image retrieval when the color and texture
features are combined. The similarity measures based on which matches are made and images are
retrieved are also discussed. For effective indexing and fast searching of images based on visual features,
neural network based pattern learning can be used to achieve effective classification.
COLOUR BASED IMAGE SEGMENTATION USING HYBRID KMEANS WITH WATERSHED SEGMENTATIONIAEME Publication
Image processing, arbitrarily manipulating an image to achieve an aesthetic standard or to support a preferred reality. The objective of segmentation is partitioning an image into distinct regions containing each pixels with similar attributes. Image segmentation can be done using thresholding, color space segmentation, k-means clustering.
Segmentation is the low-level operation concerned with partitioning images by determining disjoint and homogeneous regions or, equivalently, by finding edges or boundaries. The homogeneous regions, or the edges, are supposed to correspond, actual objects, or parts of them, within the images. Thus, in a large number of applications in image processing and computer vision, segmentation plays a fundamental role as the first step before applying to images higher-level operations such as recognition, semantic interpretation, and representation. Until very recently, attention has been focused on segmentation of gray-level images since these have been the only kind of visual information that acquisition devices were able to take the computer resources to handle. Nowadays, color image has definitely displaced monochromatic information and computation power is no longer a limitation in processing large volumes of data. In this paper proposed hybrid k-means with watershed segmentation algorithm is used segment the images. Filtering techniques is used as noise filtration method to improve the results and PSNR, MSE performance parameters has been calculated and shows the level of accuracy
An Automatic Color Feature Vector Classification Based on Clustering MethodRSIS International
In computer vision application, visual features such as
shape, color and texture are extracted to characterize images.
Each of the features is represented using one or more feature
descriptors. One of the important requirements in image
retrieval, indexing, classification, clustering, etc. is extracting
efficient features from images. The color feature is one of the
most widely used visual features. Use of color histogram is the
most common way for representing color feature. One of
disadvantage of the color histogram is that it does not take the
color spatial distribution into consideration. In this paper an
automatic color feature vector classification based on clustering
approach is presented, which effectively describes the spatial
information of color features. The image retrieval results are
compare to improved color feature vector show the acceptable
efficiency of this approach. It propose an automatic color feature
vector classification of satellite images using clustering approach.
The intention is to study cluster a set of satellite images in several
categories on the color similarity basis. The images are processed
using LAB color space in the feature extraction stage. The
resulted color-based feature vectors are clustered using an
automatic unsupervised classification algorithm. Some
experiments based on the proposed recognition technique have
also been performed. More research, however, is needed to
identify and reduce uncertainties in the image processing chain
to improve classification accuracy. The mathematical training
and prediction analysis of a general familiarity with satellite
classifications meet typical map accuracy standards.
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...CSCJournals
The document describes an image segmentation algorithm that uses both color and depth features extracted from RGBD images captured by a Kinect sensor. The algorithm clusters pixels into segments based on their color, texture, 3D spatial coordinates, surface normals, and the output of a graph-based segmentation algorithm. Depth features help resolve illumination issues and occlusion that cannot be handled by color-only methods. The algorithm was tested on commercial building images and showed potential for real-time applications.
Engineering Research Publication
Best International Journals, High Impact Journals,
International Journal of Engineering & Technical Research
ISSN : 2321-0869 (O) 2454-4698 (P)
www.erpublication.org
Content Based Image Retrieval Using Dominant Color and Texture FeaturesIJMTST Journal
The purpose of this Paper is to describe our research on different feature extraction and matching techniques in designing a Content Based Image Retrieval (CBIR) system. Due to the enormous increase in image database sizes, as well as its vast deployment in various applications, the need for CBIR development arose. Content Based Image Retrieval (CBIR) is the retrieval of images based on features such as color and texture. Image retrieval using color feature cannot provide good solution for accuracy and efficiency. The most important features are Color and texture. In this paper technique used for retrieving the images based on their content namely dominant color, texture and combination of both color and texture. The technique verifies the superiority of image retrieval using multi feature than the single feature.
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
A Combined Method with automatic parameter optimization for Multi-class Image...AM Publications
Multi-class image semantic segmentation deals with many applications in consumer electronics
fields such as image editing and image retrieval. Segmentation is done by combining the top down and bottomup
segmentation. Top-Down Process can be done by Semantic Texton Forest and bottom up- process using
JSEG. These two segmentation process can be executed in a combined manner. But this cannot choose the
optimal value of JSEG parameter for each interested semantic category. Hence an automatic parameter selection
algorithm has been proposed. An automatic parameter selection technique called an automatic multilevel
thresholding algorithm using stratified sampling and PSO is used to remedy the limitations.
Image Segmentation Using Pairwise Correlation ClusteringIJERA Editor
A pairwise hypergraph based image segmentation framework is formulated in a supervised manner for various images. The image segmentation is to infer the edge label over the pairwise hypergraph by maximizing the normalized cuts. Correlation clustering which is a graph partitioning algorithm, was shown to be effective in a number of applications such as identification, clustering of documents and image segmentation.The partitioning result is derived from a algorithm to partition a pairwise graph into disjoint groups of coherent nodes. In the pairwise correlation clustering, the pairwise graph which is used in the correlation clustering is generalized to a superpixel graph where a node corresponds to a superpixel and a link between adjacent superpixels corresponds to an edge. This pairwise correlation clustering also considers the feature vector which extracts several visual cues from a superpixel, including brightness, color, texture, and shape. Significant progress in clustering has been achieved by algorithms that are based on pairwise affinities between the datasets. The experimental results are shown by calculating the typical cut and inference in an undirected graphical model and datasets.
This document presents a new framework for color image segmentation using a combination of watershed and seed region growing algorithms. It begins with an introduction to image segmentation and discusses challenges with traditional gray-scale methods when applied to color images. The document then proposes a method using automatic seed region growing integrated with the watershed algorithm. Experimental results on an input image are shown to demonstrate the segmentation process and output images. The framework is concluded to improve upon traditional gray-scale methods for segmenting the richer information in color images.
Automatic dominant region segmentation for natural imagescsandit
Image Segmentation segments an image into different homogenous regions. An efficient
semantic based image retrieval system divides the image into different regions separated by
color or texture sometimes even both. Features are extracted from the segmented regions and
are annotated automatically. Relevant images are retrieved from the database based on the
keywords of the segmented region In this paper, automatic image segmentation is proposed to
obtained dominant region of the input natural images. Dominant region are segmented and
results are obtained . Results are also recorded in comparison to JSEG algorithm
The project aims at development of efficient segmentation method for the CBIR system. Mean-shift segmentation generates a list of potential objects which are meaningful and then these objects are clustered according to a predefined similarity measure. The method was tested on benchmark data and F-Score of .30 was achieved.
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposureiosrjce
This document discusses boundary detection techniques for images. It proposes a generalized boundary detection method (Gb) that combines low-level and mid-level image representations in a single eigenvalue problem to detect boundaries. Gb achieves state-of-the-art results at low computational cost. Soft segmentation and contour grouping methods are also introduced to further improve boundary detection accuracy with minimal extra computation. The document presents outputs of Gb on sample images and concludes that Gb effectively detects boundaries in a principled manner by jointly resolving constraints from multiple image interpretation layers in closed form.
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Similar to . Color and texture-based image segmentation using the expectation-maximization algorithm and its application to content-based image retrieval
In Content-Based Image Retrieval (CBIR) systems, the visual contents of the
images in the database are took out and represented by multi-dimensional characteristic
vectors. A well known CBIR system that retrieves images by unsupervised method known
as cluster based image retrieval system. For enhancing the performance and retrieval rate
of CBIR system, we fuse the visual contents of an image. Recently, we developed two
cluster-based CBIR systems by fusing the scores of two visual contents of an image. In this
paper, we analyzed the performance of the two recommended CBIR systems at different
levels of precision using images of varying sizes and resolutions. We also compared the
performance of the recommended systems with that of the other two existing CBIR systems
namely UFM and CLUE. Experimentally, we find that the recommended systems
outperform the other two existing systems and one recommended system also comparatively
performed better in every resolution of image.
Literature Review on Content Based Image RetrievalUpekha Vandebona
This document summarizes a literature review on content-based image retrieval (CBIR). It discusses how CBIR uses computer vision techniques to automatically extract visual features from images for retrieval, unlike traditional concept-based methods that rely on metadata/text. The key visual features discussed are color, texture, and shape. A typical CBIR system architecture includes creating an image database, automatically extracting features, searching by example or semantics, and ranking results. Distance measures are used to compare image features and evaluate retrieval performance. Combining CBIR with concept-based techniques could improve image retrieval overall.
Semi-Supervised Method of Multiple Object Segmentation with a Region Labeling...sipij
Efficient and efficient multiple object segmentation is an important task in computer vision and object recognition. In this work; we address a method to effectively discover a user’s concept when multiple objects of interest are involved in content based image retrieval. The proposed method incorporate a framework for multiple object retrieval using semi-supervised method of similar region merging and flood fill which models the spatial and appearance relations among image pixels. To improve the effectiveness of similarity based region merging we propose a new similarity based object retrieval. The users only need to roughly indicate the after which steps desired objects contour is obtained during the automatic merging of similar regions. A novel similarity based region merging mechanism is proposed to guide the merging process with the help of mean shift technique and objects detection using region labeling and flood fill. A region R is merged with its adjacent regions Q if Q has highest similarity with Q (using Bhattacharyya descriptor) among all Q’s adjacent regions. The proposed method automatically merges the regions that are initially segmented through mean shift technique, and then effectively extracts the object contour by merging all similar regions. Extensive experiments are performed on 12 object classes (224 images total) show promising results.
WEB IMAGE RETRIEVAL USING CLUSTERING APPROACHEScscpconf
Image retrieval system is an active area to propose a new approach to retrieve images from the
large image database. In this concerned, we proposed an algorithm to represent images using
divisive based and partitioned based clustering approaches. The HSV color component and Haar wavelet transform is used to extract image features. These features are taken to segment an image to obtain objects. For segmenting an image, we used modified k-means clustering algorithm to group similar pixel together into K groups with cluster centers. To modify Kmeans, we proposed a divisive based clustering algorithm to determine the number of cluster and get back with number of cluster to k-means to obtain significant object groups. In addition, we also discussed the similarity distance measure using threshold value and object uniqueness to quantify the results.
Color and texture based image retrievaleSAT Journals
Abstract Content-based image retrieval (CBIR) is an vital research area for manipulating bulky image databases and records. Alongside the conventional method where the images are searched on the basis of words, CBIR system uses visual contents to retrieve the images. In content based image retrieval systems texture and color features have been the primal descriptors. We use HSV color information and mean of the image as texture information. The performance of proposed scheme is calculated on the basis of precision, recall and accuracy. As an effect, the blend of color and texture features of the image provides strong feature set for image retrieval. Keywords: image retrieval, HSV color space, color histogram, image texture.
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Ijaems apr-2016-16 Active Learning Method for Interactive Image RetrievalINFOGAIN PUBLICATION
With many possible multimedia applications, content-based image retrieval (CBIR) has recently gained more interest for image management and web search. CBIR is a technique that utilizes the visual content of an image, to search for similar images in large-scale image databases, according to a user’s concern. In image retrieval algorithms, retrieval is according to feature similarities with respect to the query, ignoring the similarities among images in database. To use the feature similarities information, this paper presents the k-means clustering algorithm to image retrieval system. This clustering algorithm optimizes the relevance results by firstly clustering the similar images in the database. In this paper, we are also implementing wavelet transform which demonstrates significant rough and precise filtering. We also apply the Euclidean distance metric and input a query image based on similarity features of which we can retrieve the output images. The results show that the proposed approach can greatly improve the efficiency and performances of image retrieval.
ER Publication,
IJETR, IJMCTR,
Journals,
International Journals,
High Impact Journals,
Monthly Journal,
Good quality Journals,
Research,
Research Papers,
Research Article,
Free Journals, Open access Journals,
erpublication.org,
Engineering Journal,
Science Journals,
SEMANTIC IMAGE RETRIEVAL USING MULTIPLE FEATUREScscpconf
In Content Based Image Retrieval (CBIR) some problem such as recognizing the similar
images, the need for databases, the semantic gap, and retrieving the desired images from huge
collections are the keys to improve. CBIR system analyzes the image content for indexing,
management, extraction and retrieval via low-level features such as color, texture and shape.
To achieve higher semantic performance, recent system seeks to combine the low-level features
of images with high-level features that conation perceptual information for human beings.
Performance improvements of indexing and retrieval play an important role for providing
advanced CBIR services. To overcome these above problems, a new query-by-image technique
using combination of multiple features is proposed. The proposed technique efficiently sifts through the dataset of images to retrieve semantically similar images.
The efficiency and quality of a feature descriptor are critical to the user experience of many computer vision applications. However, the existing descriptors are either too computationally expensive to achieve real-time performance, or not sufficiently distinctive to identify correct matches from a large database with various transformations. In this paper, we propose a highly efficient and distinctive binary descriptor, called local difference binary (LDB). LDB directly computes a binary string for an image patch using simple intensity and gradient difference tests on pair wise grid cells within the patch. A multiple-gridding strategy and a salient bit-selection method are applied to capture the distinct patterns of the patch at different spatial granularities. Experimental results demonstrate that compared to the existing state-of-the-art binary descriptors, primarily designed for speed, LDB has similar construction efficiency, while achieving a greater accuracy and faster speed for mobile object recognition and tracking tasks.
Image Retrieval using Equalized Histogram Image Bins MomentsIDES Editor
CBIR operates on a totally different principle
from keyword indexing. Primitive features characterizing
image content, such as color, texture, and shape are computed
for both stored and query images, and used to identify the
images most closely matching the query. There have been
many approaches to decide and extract the features of images
in the database. Towards this goal we propose a technique by
which the color content of images is automatically extracted to
form a class of meta-data that is easily indexed. The color
indexing algorithm uses the back-projection of binary color
sets to extract color regions from images. This technique use
without histogram of image histogram bins of red, green and
blue color. The feature vector is composed of mean, standard
deviation and variance of 16 histogram bins of each color
space. The new proposed methods are tested on the database
of 600 images and the results are in the form of precision and
recall.
Content based image retrieval based on shape with texture featuresAlexander Decker
This document describes a content-based image retrieval system that extracts shape and texture features from images. It uses the HSV color space and wavelet transform for feature extraction. Color features are extracted by quantizing the H, S, and V components of HSV into unequal intervals based on human color perception. Texture features are extracted using wavelet transforms. The color and texture features are then combined to form a feature vector for each image. During retrieval, the similarity between a query image and images in the database is measured using the Euclidean distance between their feature vectors. The results show that retrieving images using HSV color features provides more accurate results and faster retrieval times compared to using RGB color features.
Texture Segmentation Based on Multifractal Dimensionijsc
Texture segmentation can be considered the most important problem, since human can distinguish different
textures quit easily, but the automatic segmentation is quit complex and it is still an open problem for
research. In this paper focus on implement novel supervised algorithm for multitexture segmentation and
this algorithm based on blocking procedure where each image divide into block (16×16 pixels) and extract
vector feature for each block to classification these block based on these feature. These feature extract
using Box Counting Method (BCM). BCM generate single feature for each block and this feature not
enough to characterize each block ,therefore, must be implement algorithm provide more than one slide for
the image based on new method produce multithresolding, after this use BCM to generate single feature for
each slide.
Texture Segmentation Based on Multifractal Dimension ijsc
This document presents a new texture segmentation algorithm based on multifractal dimension. The algorithm divides an image into blocks and extracts feature vectors for each block using box counting method on multiple thresholds of the image. A supervised learning phase is used to classify blocks based on these feature vectors by extracting mean and standard deviation values for sample windows labeled by an expert. The algorithm was tested on multi-texture images by extracting feature vectors for each small block and classifying them based on the trained classifier.
A Hybrid Approach for Content Based Image Retrieval SystemIOSR Journals
This document describes a hybrid approach for content-based image retrieval. It combines several spatial features - row sum, column sum, forward and backward diagonal sums - and histograms to represent images with feature vectors. Euclidean distance is used to calculate similarity between a query image's feature vector and those in the database. The approach is evaluated using precision-recall calculations on different image groups, showing the hybrid method performs best by combining multiple features.
A Novel Method for Content Based Image Retrieval using Local Features and SVM...IRJET Journal
1) The document presents a novel approach for content-based image retrieval that uses local features like color, texture, and edges extracted from images.
2) It extracts these features and uses an SVM classifier to optimize retrieval results. This improves accuracy compared to other techniques that use only one content feature.
3) The proposed system is tested on parameters like accuracy, sensitivity, specificity, error rate, and retrieval time, and shows better performance than other methods.
Performance Evaluation Of Ontology And Fuzzybase Cbiracijjournal
In This Paper, We Have Done Performance Evaluation Of Ontology Using Low-Level Features Like
Color, Texture And Shape Based Cbir, With Topic Specific Cbir.The Resulting Ontology Can Be Used
To Extract The Appropriate Images From The Image Database. Retrieving Appropriate Images From An
Image Database Is One Of The Difficult Tasks In Multimedia Technology. Our Results Show That The
Values Of Recall And Precision Can Be Enhanced And This Also Shows That Semantic Gap Can Also Be
Reduced. The Proposed Algorithm Also Extracts The Texture Values From The Images Automatically
With Also Its Category (Like Smooth, Course Etc) As Well As Its Technical Interpretation
PERFORMANCE EVALUATION OF ONTOLOGY AND FUZZYBASE CBIRacijjournal
IN THIS PAPER, WE HAVE DONE PERFORMANCE EVALUATION OF ONTOLOGY USING LOW-LEVEL FEATURES LIKE
COLOR, TEXTURE AND SHAPE BASED CBIR, WITH TOPIC SPECIFIC CBIR.THE RESULTING ONTOLOGY CAN BE USED
TO EXTRACT THE APPROPRIATE IMAGES FROM THE IMAGE DATABASE. RETRIEVING APPROPRIATE IMAGES FROM AN
IMAGE DATABASE IS ONE OF THE DIFFICULT TASKS IN MULTIMEDIA TECHNOLOGY. OUR RESULTS SHOW THAT THE
VALUES OF RECALL AND PRECISION CAN BE ENHANCED AND THIS ALSO SHOWS THAT SEMANTIC GAP CAN ALSO BE
REDUCED. THE PROPOSED ALGORITHM ALSO EXTRACTS THE TEXTURE VALUES FROM THE IMAGES AUTOMATICALLY
WITH ALSO ITS CATEGORY (LIKE SMOOTH, COURSE ETC) AS WELL AS ITS TECHNICAL INTERPRETATION.
Global Descriptor Attributes Based Content Based Image Retrieval of Query ImagesIJERA Editor
The need for efficient content-based image retrieval system has increased hugely. Efficient and effective retrieval techniques of images are desired because of the explosive growth of digital images. Content based image retrieval (CBIR) is a promising approach because of its automatic indexing retrieval based on their semantic features and visual appearance. In this proposed system we investigate method for describing the contents of images which characterizes images by global descriptor attributes, where global features are extracted to make system more efficient by using color features which are color expectancy, color variance, skewness and texture feature correlation.
Research Inventy : International Journal of Engineering and Scienceinventy
Research Inventy : International Journal of Engineering and Science is published by the group of young academic and industrial researchers with 12 Issues per year. It is an online as well as print version open access journal that provides rapid publication (monthly) of articles in all areas of the subject such as: civil, mechanical, chemical, electronic and computer engineering as well as production and information technology. The Journal welcomes the submission of manuscripts that meet the general criteria of significance and scientific excellence. Papers will be published by rapid process within 20 days after acceptance and peer review process takes only 7 days. All articles published in Research Inventy will be peer-reviewed.
Similar to . Color and texture-based image segmentation using the expectation-maximization algorithm and its application to content-based image retrieval (20)
Object segmentation by alignment of poselet activations to image contoursirisshicat
This document proposes techniques to segment objects in images using a combination of bottom-up (image edges) and top-down (object detector) cues. The key techniques are:
1. Extending an existing poselet detector to 19 additional categories beyond people. Poselets can predict masks for parts of objects.
2. Non-rigidly aligning the predicted poselet masks to image contours to increase segmentation precision and remove false positives.
3. Spatially smoothing the aligned masks while ensuring non-overlapping object regions using a variational technique.
4. Refining the segmentation further based on self-similarity of small image patches. The approach achieves competitive results on the PASCAL VOC benchmark
1) Biased Normalized Cuts presents a modification of Normalized Cuts that incorporates priors to allow for constrained image segmentation.
2) It seeks solutions that are sufficiently "correlated" with noisy top-down priors, like an object detector, and can be computed quickly given the unconstrained solution.
3) The algorithm constructs a "biased normalized cut vector" that linearly combines eigenvectors such that those correlated with a user-specified seed vector are upweighted while inversely correlated ones have their sign flipped.
A probabilistic model for recursive factorized image features pptirisshicat
The document describes a probabilistic model called recursive latent Dirichlet allocation (rLDA) for hierarchical image modeling. rLDA is based on latent Dirichlet allocation and has multiple layers of representations with increasing spatial support, where each layer learns representations jointly across layers through joint inference. This allows for distributed coding of local image features in a hierarchical manner while performing full Bayesian inference. The model is evaluated for its ability to learn hierarchical representations from images.
A probabilistic model for recursive factorized image featuresirisshicat
This document proposes a probabilistic model for learning hierarchical visual representations in a recursive manner. The model, based on Latent Dirichlet Allocation, learns image features at multiple layers of abstraction jointly rather than in a strictly feedforward way. The model represents local image patches as distributions over visual words at the lowest layer, and higher layers learn distributions over the representations of lower layers. Evaluating the model on a standard recognition dataset, it outperforms existing hierarchical models and achieves performance on par with state-of-the-art single-feature models, demonstrating the benefits of joint learning and inference in hierarchical visual processing.
The document discusses the mean shift algorithm, a non-parametric technique for analyzing complex multimodal feature spaces and estimating the stationary points (modes) of the underlying probability density function without explicitly estimating it. It provides an intuitive description of mean shift using a distribution of billiard balls, and outlines how mean shift uses kernel density estimation to perform gradient ascent and converge at the densest regions, allowing it to be used for tasks like mode detection, clustering, and image segmentation.
The mean shift procedure is a general nonparametric technique for analyzing complex multimodal feature spaces and delineating arbitrarily shaped clusters. It works by recursively finding the nearest stationary point of the underlying density function, which corresponds to the mode of the density. The mean shift procedure relates to kernel density estimation and robust M-estimators of location. It provides a versatile tool for feature space analysis that can solve many low-level computer vision tasks with few parameters.
Shape matching and object recognition using shape context belongie pami02irisshicat
1) The document presents a novel approach for measuring shape similarity and using it for object recognition. It involves finding point correspondences between shapes, estimating an aligning transformation, and computing distance as a sum of matching errors and transformation magnitude.
2) At the core is using a "shape context" descriptor at sample points to solve the correspondence problem as a graph matching problem. This provides correspondences to estimate an aligning transformation.
3) Shape similarity is then a measure of matching errors between corresponding points after alignment, allowing nearest neighbor classification for recognition. Results are shown for various datasets.
Shape matching and object recognition using shape contextsirisshicat
The document discusses a seminar on shape matching and object recognition using shape contexts. Shape contexts are shape descriptors that can be used to match shapes. They involve creating log polar histograms of shapes and finding correspondences between points on different shapes using bipartite graph matching. The approach was used for applications like digit recognition, achieving an error rate of 63% using 20,000 training examples, and 3D object detection using 72 views per object.
Dandelion Hashtable: beyond billion requests per second on a commodity serverAntonios Katsarakis
This slide deck presents DLHT, a concurrent in-memory hashtable. Despite efforts to optimize hashtables, that go as far as sacrificing core functionality, state-of-the-art designs still incur multiple memory accesses per request and block request processing in three cases. First, most hashtables block while waiting for data to be retrieved from memory. Second, open-addressing designs, which represent the current state-of-the-art, either cannot free index slots on deletes or must block all requests to do so. Third, index resizes block every request until all objects are copied to the new index. Defying folklore wisdom, DLHT forgoes open-addressing and adopts a fully-featured and memory-aware closed-addressing design based on bounded cache-line-chaining. This design offers lock-free index operations and deletes that free slots instantly, (2) completes most requests with a single memory access, (3) utilizes software prefetching to hide memory latencies, and (4) employs a novel non-blocking and parallel resizing. In a commodity server and a memory-resident workload, DLHT surpasses 1.6B requests per second and provides 3.5x (12x) the throughput of the state-of-the-art closed-addressing (open-addressing) resizable hashtable on Gets (Deletes).
FREE A4 Cyber Security Awareness Posters-Social Engineering part 3Data Hops
Free A4 downloadable and printable Cyber Security, Social Engineering Safety and security Training Posters . Promote security awareness in the home or workplace. Lock them Out From training providers datahops.com
Monitoring and Managing Anomaly Detection on OpenShift.pdfTosin Akinosho
Monitoring and Managing Anomaly Detection on OpenShift
Overview
Dive into the world of anomaly detection on edge devices with our comprehensive hands-on tutorial. This SlideShare presentation will guide you through the entire process, from data collection and model training to edge deployment and real-time monitoring. Perfect for those looking to implement robust anomaly detection systems on resource-constrained IoT/edge devices.
Key Topics Covered
1. Introduction to Anomaly Detection
- Understand the fundamentals of anomaly detection and its importance in identifying unusual behavior or failures in systems.
2. Understanding Edge (IoT)
- Learn about edge computing and IoT, and how they enable real-time data processing and decision-making at the source.
3. What is ArgoCD?
- Discover ArgoCD, a declarative, GitOps continuous delivery tool for Kubernetes, and its role in deploying applications on edge devices.
4. Deployment Using ArgoCD for Edge Devices
- Step-by-step guide on deploying anomaly detection models on edge devices using ArgoCD.
5. Introduction to Apache Kafka and S3
- Explore Apache Kafka for real-time data streaming and Amazon S3 for scalable storage solutions.
6. Viewing Kafka Messages in the Data Lake
- Learn how to view and analyze Kafka messages stored in a data lake for better insights.
7. What is Prometheus?
- Get to know Prometheus, an open-source monitoring and alerting toolkit, and its application in monitoring edge devices.
8. Monitoring Application Metrics with Prometheus
- Detailed instructions on setting up Prometheus to monitor the performance and health of your anomaly detection system.
9. What is Camel K?
- Introduction to Camel K, a lightweight integration framework built on Apache Camel, designed for Kubernetes.
10. Configuring Camel K Integrations for Data Pipelines
- Learn how to configure Camel K for seamless data pipeline integrations in your anomaly detection workflow.
11. What is a Jupyter Notebook?
- Overview of Jupyter Notebooks, an open-source web application for creating and sharing documents with live code, equations, visualizations, and narrative text.
12. Jupyter Notebooks with Code Examples
- Hands-on examples and code snippets in Jupyter Notebooks to help you implement and test anomaly detection models.
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc
How does your privacy program stack up against your peers? What challenges are privacy teams tackling and prioritizing in 2024?
In the fifth annual Global Privacy Benchmarks Survey, we asked over 1,800 global privacy professionals and business executives to share their perspectives on the current state of privacy inside and outside of their organizations. This year’s report focused on emerging areas of importance for privacy and compliance professionals, including considerations and implications of Artificial Intelligence (AI) technologies, building brand trust, and different approaches for achieving higher privacy competence scores.
See how organizational priorities and strategic approaches to data security and privacy are evolving around the globe.
This webinar will review:
- The top 10 privacy insights from the fifth annual Global Privacy Benchmarks Survey
- The top challenges for privacy leaders, practitioners, and organizations in 2024
- Key themes to consider in developing and maintaining your privacy program
This presentation provides valuable insights into effective cost-saving techniques on AWS. Learn how to optimize your AWS resources by rightsizing, increasing elasticity, picking the right storage class, and choosing the best pricing model. Additionally, discover essential governance mechanisms to ensure continuous cost efficiency. Whether you are new to AWS or an experienced user, this presentation provides clear and practical tips to help you reduce your cloud costs and get the most out of your budget.
Trusted Execution Environment for Decentralized Process MiningLucaBarbaro3
Presentation of the paper "Trusted Execution Environment for Decentralized Process Mining" given during the CAiSE 2024 Conference in Cyprus on June 7, 2024.
Fueling AI with Great Data with Airbyte WebinarZilliz
This talk will focus on how to collect data from a variety of sources, leveraging this data for RAG and other GenAI use cases, and finally charting your course to productionalization.
Digital Banking in the Cloud: How Citizens Bank Unlocked Their MainframePrecisely
Inconsistent user experience and siloed data, high costs, and changing customer expectations – Citizens Bank was experiencing these challenges while it was attempting to deliver a superior digital banking experience for its clients. Its core banking applications run on the mainframe and Citizens was using legacy utilities to get the critical mainframe data to feed customer-facing channels, like call centers, web, and mobile. Ultimately, this led to higher operating costs (MIPS), delayed response times, and longer time to market.
Ever-changing customer expectations demand more modern digital experiences, and the bank needed to find a solution that could provide real-time data to its customer channels with low latency and operating costs. Join this session to learn how Citizens is leveraging Precisely to replicate mainframe data to its customer channels and deliver on their “modern digital bank” experiences.
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...Alex Pruden
Folding is a recent technique for building efficient recursive SNARKs. Several elegant folding protocols have been proposed, such as Nova, Supernova, Hypernova, Protostar, and others. However, all of them rely on an additively homomorphic commitment scheme based on discrete log, and are therefore not post-quantum secure. In this work we present LatticeFold, the first lattice-based folding protocol based on the Module SIS problem. This folding protocol naturally leads to an efficient recursive lattice-based SNARK and an efficient PCD scheme. LatticeFold supports folding low-degree relations, such as R1CS, as well as high-degree relations, such as CCS. The key challenge is to construct a secure folding protocol that works with the Ajtai commitment scheme. The difficulty, is ensuring that extracted witnesses are low norm through many rounds of folding. We present a novel technique using the sumcheck protocol to ensure that extracted witnesses are always low norm no matter how many rounds of folding are used. Our evaluation of the final proof system suggests that it is as performant as Hypernova, while providing post-quantum security.
Paper Link: https://eprint.iacr.org/2024/257
In the realm of cybersecurity, offensive security practices act as a critical shield. By simulating real-world attacks in a controlled environment, these techniques expose vulnerabilities before malicious actors can exploit them. This proactive approach allows manufacturers to identify and fix weaknesses, significantly enhancing system security.
This presentation delves into the development of a system designed to mimic Galileo's Open Service signal using software-defined radio (SDR) technology. We'll begin with a foundational overview of both Global Navigation Satellite Systems (GNSS) and the intricacies of digital signal processing.
The presentation culminates in a live demonstration. We'll showcase the manipulation of Galileo's Open Service pilot signal, simulating an attack on various software and hardware systems. This practical demonstration serves to highlight the potential consequences of unaddressed vulnerabilities, emphasizing the importance of offensive security practices in safeguarding critical infrastructure.
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slackshyamraj55
Discover the seamless integration of RPA (Robotic Process Automation), COMPOSER, and APM with AWS IDP enhanced with Slack notifications. Explore how these technologies converge to streamline workflows, optimize performance, and ensure secure access, all while leveraging the power of AWS IDP and real-time communication via Slack notifications.
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUpanagenda
Webinar Recording: https://www.panagenda.com/webinars/hcl-notes-und-domino-lizenzkostenreduzierung-in-der-welt-von-dlau/
DLAU und die Lizenzen nach dem CCB- und CCX-Modell sind für viele in der HCL-Community seit letztem Jahr ein heißes Thema. Als Notes- oder Domino-Kunde haben Sie vielleicht mit unerwartet hohen Benutzerzahlen und Lizenzgebühren zu kämpfen. Sie fragen sich vielleicht, wie diese neue Art der Lizenzierung funktioniert und welchen Nutzen sie Ihnen bringt. Vor allem wollen Sie sicherlich Ihr Budget einhalten und Kosten sparen, wo immer möglich. Das verstehen wir und wir möchten Ihnen dabei helfen!
Wir erklären Ihnen, wie Sie häufige Konfigurationsprobleme lösen können, die dazu führen können, dass mehr Benutzer gezählt werden als nötig, und wie Sie überflüssige oder ungenutzte Konten identifizieren und entfernen können, um Geld zu sparen. Es gibt auch einige Ansätze, die zu unnötigen Ausgaben führen können, z. B. wenn ein Personendokument anstelle eines Mail-Ins für geteilte Mailboxen verwendet wird. Wir zeigen Ihnen solche Fälle und deren Lösungen. Und natürlich erklären wir Ihnen das neue Lizenzmodell.
Nehmen Sie an diesem Webinar teil, bei dem HCL-Ambassador Marc Thomas und Gastredner Franz Walder Ihnen diese neue Welt näherbringen. Es vermittelt Ihnen die Tools und das Know-how, um den Überblick zu bewahren. Sie werden in der Lage sein, Ihre Kosten durch eine optimierte Domino-Konfiguration zu reduzieren und auch in Zukunft gering zu halten.
Diese Themen werden behandelt
- Reduzierung der Lizenzkosten durch Auffinden und Beheben von Fehlkonfigurationen und überflüssigen Konten
- Wie funktionieren CCB- und CCX-Lizenzen wirklich?
- Verstehen des DLAU-Tools und wie man es am besten nutzt
- Tipps für häufige Problembereiche, wie z. B. Team-Postfächer, Funktions-/Testbenutzer usw.
- Praxisbeispiele und Best Practices zum sofortigen Umsetzen
A Comprehensive Guide to DeFi Development Services in 2024Intelisync
DeFi represents a paradigm shift in the financial industry. Instead of relying on traditional, centralized institutions like banks, DeFi leverages blockchain technology to create a decentralized network of financial services. This means that financial transactions can occur directly between parties, without intermediaries, using smart contracts on platforms like Ethereum.
In 2024, we are witnessing an explosion of new DeFi projects and protocols, each pushing the boundaries of what’s possible in finance.
In summary, DeFi in 2024 is not just a trend; it’s a revolution that democratizes finance, enhances security and transparency, and fosters continuous innovation. As we proceed through this presentation, we'll explore the various components and services of DeFi in detail, shedding light on how they are transforming the financial landscape.
At Intelisync, we specialize in providing comprehensive DeFi development services tailored to meet the unique needs of our clients. From smart contract development to dApp creation and security audits, we ensure that your DeFi project is built with innovation, security, and scalability in mind. Trust Intelisync to guide you through the intricate landscape of decentralized finance and unlock the full potential of blockchain technology.
Ready to take your DeFi project to the next level? Partner with Intelisync for expert DeFi development services today!
Introduction of Cybersecurity with OSS at Code Europe 2024Hiroshi SHIBATA
I develop the Ruby programming language, RubyGems, and Bundler, which are package managers for Ruby. Today, I will introduce how to enhance the security of your application using open-source software (OSS) examples from Ruby and RubyGems.
The first topic is CVE (Common Vulnerabilities and Exposures). I have published CVEs many times. But what exactly is a CVE? I'll provide a basic understanding of CVEs and explain how to detect and handle vulnerabilities in OSS.
Next, let's discuss package managers. Package managers play a critical role in the OSS ecosystem. I'll explain how to manage library dependencies in your application.
I'll share insights into how the Ruby and RubyGems core team works to keep our ecosystem safe. By the end of this talk, you'll have a better understanding of how to safeguard your code.
For the full video of this presentation, please visit: https://www.edge-ai-vision.com/2024/06/temporal-event-neural-networks-a-more-efficient-alternative-to-the-transformer-a-presentation-from-brainchip/
Chris Jones, Director of Product Management at BrainChip , presents the “Temporal Event Neural Networks: A More Efficient Alternative to the Transformer” tutorial at the May 2024 Embedded Vision Summit.
The expansion of AI services necessitates enhanced computational capabilities on edge devices. Temporal Event Neural Networks (TENNs), developed by BrainChip, represent a novel and highly efficient state-space network. TENNs demonstrate exceptional proficiency in handling multi-dimensional streaming data, facilitating advancements in object detection, action recognition, speech enhancement and language model/sequence generation. Through the utilization of polynomial-based continuous convolutions, TENNs streamline models, expedite training processes and significantly diminish memory requirements, achieving notable reductions of up to 50x in parameters and 5,000x in energy consumption compared to prevailing methodologies like transformers.
Integration with BrainChip’s Akida neuromorphic hardware IP further enhances TENNs’ capabilities, enabling the realization of highly capable, portable and passively cooled edge devices. This presentation delves into the technical innovations underlying TENNs, presents real-world benchmarks, and elucidates how this cutting-edge approach is positioned to revolutionize edge AI across diverse applications.
“Temporal Event Neural Networks: A More Efficient Alternative to the Transfor...
. Color and texture-based image segmentation using the expectation-maximization algorithm and its application to content-based image retrieval
1. Color- and Texture-Based Image Segmentation Using EM
and Its Application to Content-Based Image Retrieval
Serge Belongie, Chad Carson, Hayit Greenspan, and Jitendra Malik
Computer Science Division
University of California at Berkeley
Berkeley, CA 94720
fsjb,carson,hayit,malikg@cs.berkeley.edu
Abstract their low-level features (“stuff”), with little regard for
the spatial organization of those features.
Retrieving images from large and varied collections using
image content as a key is a challenging and important prob- Systems based on user querying are often unintuitive
lem. In this paper we present a new image representation and offer little help in understanding why certain im-
which provides a transformation from the raw pixel data to ages were returned and how to refine the query. Often
a small set of image regions which are coherent in color and the user knows only that he has submitted a query for,
texture space. This so-called “blobworld” representation is say, a bear and retrieved very few pictures of bears in
based on segmentation using the Expectation-Maximization return.
algorithm on combined color and texture features. The tex-
ture features we use for the segmentation arise from a new For general image collections, there are currently no
approach to texture description and scale selection. systems that automatically classify images or recog-
We describe a system that uses the blobworld representa- nize the objects they contain.
tion to retrieve images. An important and unique aspect of
the system is that, in the context of similarity-based query- In this paper we present a new image representation,
ing, the user is allowed to view the internal representation “blobworld,” and a retrieval system based on this repre-
of the submitted image and the query results. Similar sys- sentation. While blobworld does not exist completely in
tems do not offer the user this view into the workings of the the “thing” domain, it recognizes the nature of images as
system; consequently, the outcome of many queries on these combinations of objects, and both querying and learning in
systems can be quite inexplicable, despite the availability of blobworld are more meaningful than they are with simple
knobs for adjusting the similarity metric. “stuff” representations.
We use the Expectation-Maximization (EM) algorithm to
perform automatic segmentation based on image features.
1 Introduction EM iteratively models the joint distribution of color and
texture with a mixture of Gaussians; the resulting pixel-
Very large collections of images are growing ever more cluster memberships provide a segmentation of the image.
common. From stock photo collections to proprietary After the image is segmented into regions, a description
databases to the Web, these collections are diverse and often of each region’s color, texture, and spatial characteristics is
poorly indexed; unfortunately, image retrieval systems have produced. In a querying task, the user can access the regions
not kept pace with the collections they are searching. The directly, in order to see the segmentation of the query image
shortcomings of these systems are due both to the image and specify which aspects of the image are important to the
representations they use and to their methods of accessing query. When query results are returned, the user sees the
those representations to find images: blobworld representation of the returned images; this assists
greatly in refining the query.
While users generally want to find images containing
particular objects (“things”) [4, 6], most existing im- We begin this paper by briefly discussing the current state
age retrieval systems represent images based only on of image retrieval. In Section 2 we describe the blobworld
representation, from features through segmentation to region
To appear at ICCV ’98. Copyright (c) 1998 IEEE. description. In Section 3 we present a query system based
2. on blobworld, as well as results from queries in a collection sents a set of pixels that are coherent in their color and local
of highly varied natural images. texture properties; the motivation is to reduce the amount
of raw data presented by the image while preserving the in-
1.1 Background formation needed for the image understanding task. Given
Current image database systems include IBM’s Query by the unconstrained nature of the images in our database, it is
Image Content (QBIC) [18], Photobook [20], Virage [10], important that the tools we employ to meet this goal be as
Candid [14], and Chabot [19]. These systems primarily use general as possible without sacrificing an undue amount of
low-level image properties; several of them include some de- descriptive power.
gree of automatic segmentation. None of the systems codes
spatial organization in a way that supports object queries. 2.1.1 Color
Classical object recognition techniques usually rely on
clean segmentation of the object from the rest of the image Color is a very important cue in extracting informa-
or are designed for fixed geometric objects such as machine tion from images. Color histograms are commonly used
parts. Neither constraint holds in our case: the shape, size, in content-based retrieval systems [18, 19, 24] and have
and color of objects like cheetahs and polar bears are quite proven to be very useful; however, the global characteriza-
variable, and segmentation is imperfect. Clearly, classical tion is poor at, for example, distinguishing between a field
object recognition does not apply. More recent techniques of orange flowers and a tiger, because it lacks information
[21] can identify specific objects drawn from a finite (on about how the color is distributed spatially. It is important
the order of 100) collection, but no present technique is to group color in localized regions and to fuse color with
effective at the general image analysis task, which requires textural properties.
both image segmentation and image classification. We treat the hue-saturation-value (HSV) color space as
Promising work by Lipson et al. [16] retrieves images a cone: for a given point h; s; v, h and sv are the angular
based on spatial and photometric relationships within and and radial coordinates of the point on a disk of radius v at
across image regions. Little or no segmentation is done; the height v; all coordinates range from 0 to 1. Points with
regions are derived from low-resolution images. small v are black, regardless of their h and s values. The
Earlier work has used the EM algorithm and/or the Min- cone representation maps all such points to the apex of the
imum Description Length (MDL) principle to perform seg- cone, so they are close to one another. The Cartesian coor-
mentation based on motion [1, 25] or scaled intensities [26], dinates of points in the cone, sv cos2h; sv sin2h; v,
but EM has not previously been used on joint color and can now be used to find color differences. This encoding
texture. Related approaches such as deterministic anneal- allows us to operationalize the fact that hue differences are
ing [11] and classical clustering [12] have been applied to meaningless for very small saturations (those near the cone’s
texture segmentation without color. axis). However, it ignores the fact that for large values and
saturations, hue differences are perceptually more relevant
than saturation and value differences.
2 The blobworld image representation
The blobworld representation is related to the notion of
2.1.2 Texture
photographic or artistic scene composition. In the sense
discussed in [23], the blobworld descriptors constitute an Texture is a well-researched property of image regions,
example of a summary representation because they are con- and many texture descriptors have been proposed, including
cise and relatively easy to process in a querying framework. multi-orientation filter banks [17] and the second-moment
Blobworld is distinct from color-layout matching as in matrix [5, 8]. We will not elaborate here on the classical
QBIC in that it is designed to find objects or parts of ob- approaches to texture segmentation and classification, both
jects. Each image may be visualized by an ensemble of 2-D of which are challenging and well-studied tasks. Rather, we
ellipses, or “blobs,” each of which possesses a number of introduce a new perspective related to texture descriptors and
attributes. The number of blobs in an image is typically less texture grouping motivated by the content-based retrieval
than ten. Each blob represents a region of the image which task.
is roughly homogeneous with respect to color or texture. While color is a point property, texture is a local-
A blob is described by its dominant colors, mean texture neighborhood property. It does not make sense to talk about
descriptors, and spatial centroid and scatter matrix. (See the texture of zebra stripes at a particular pixel without spec-
Figs. 3–4 for a visualization of blobworld.) ifying a neighborhood around that pixel. In order for a
texture descriptor to be useful, it must provide an adequate
2.1 Extracting color and texture features description of the underlying texture parameters and it must
Our goal is to assign the pixels in the original image to a be computed in a neighborhood which is appropriate to the
relatively small number of groups, where each group repre- local structure being described.
2
3. with the raw entries in M , it is more common to deal with
its eigenstructure [2, 5]. Consider a fixed scale and pixel
e a c d location, let 1 and 2 (1 2 ) denote the eigenvalues of
M at that location, and let denote the argument of the
principal eigenvector of M . When 1 is large compared to
b 2 , the local neighborhood possesses a dominant orientation,
as specified by . When the eigenvalues are comparable,
there is no preferred orientation, and when both eigenval-
ues are negligible, the local neighborhood is approximately
constant.
Scale selection
We may think of as controlling the size of the integra-
(a) flow; (b) flow; (c) 2-D texture; (d) edge (e) uniform tion window around each pixel within which the outer prod-
σ = 1.5 σ = 2.5 σ = 1.5 σ=0 σ=0 uct of the gradient vectors is averaged. has been called
the integration scale or artificial scale by various authors
Figure 1. Five sample patches from a zebra im- [5, 8] to distinguish it from the natural scale used in linear
age. a and b have stripes 1-D ow of dif- smoothing of raw image intensities. Note that = x; y;
ferent scales and orientations, c is a region of 2-D the scale varies across the image.1
texture, d contains an edge, and e is a uniform In order to select the scale at which M is computed,
region. i.e. to determine the function x; y, we make use of a
local image property known as polarity.2 The polarity is
The first requirement could be met to an arbitrary degree a measure of the extent to which the gradient vectors in a
of satisfaction by using multi-orientation filter banks such as certain neighborhood all point in the same direction. (In the
steerable filters; we chose a simpler method that is sufficient computation of second moments, this information is lost in
for our purposes. The second requirement, which may be the outer product operation; i.e., gradient vector directions
thought of as the problem of scale selection, does not enjoy differing by 180 are indistinguishable.) The polarity at a
the same level of attention in the literature. This is unfortu- given pixel is computed with respect to the dominant ori-
nate, since texture descriptors computed at the wrong scale entation in the neighborhood of that pixel. For ease of
only confuse the issue. notation, let us consider a fixed scale and pixel location. We
In this work, we introduce a novel method of scale selec- define polarity as
tion which works in tandem with a fairly simple but informa-
tive set of texture descriptors. The scale selection method is p=
jE+ , E, j
based on edge/bar polarity stabilization, and the texture de- E+ + E,
scriptors arise from the windowed second moment matrix. The definitions of E+ and E, are
Both are derived from the gradient of the image intensity, X
which we denote by rI . We compute rI using the first E+ = G x; y rI n +
ˆ
difference approximation along each dimension. This oper- 2
x;y Ω
ation is often accompanied by smoothing, but we have found
this preprocessing operation unnecessary for the images in and X
our collection. E, = G x; y rI n ,
ˆ
To make the notion of scale concrete, we define the scale 2
x;y Ω
to be the width of the Gaussian window within which the where q + and q , are the rectified positive and negative
gradient vectors of the image are pooled. The second mo- parts of their argument, n is a unit vector perpendicular to ,
ˆ
ment matrix for the vectors within this window, computed and Ω represents the neighborhood under consideration. We
about each pixel in the image, can be approximated using can think of E+ and E, as measures of how many gradient
vectors in Ω are on the “positive side” and “negative side”
M x; y = G x; y rI rI T (1)
of the dominant orientation, respectively. Note that p ranges
where G x; y is a separable binomial approximation to a from 0 to 1. A similar measure is used in [15] to distinguish
Gaussian smoothing kernel with variance 2 . a flow pattern from an edge.
At each pixel location, M x; y is a 2 2 symmetric 1 Strictly speaking, eqn. (1) is a sliding inner product, not a convolution,
positive semidefinite matrix; thus it provides us with three since x; y is spatially variant.
pieces of information about each pixel. Rather than work 2 Polarity is related to the quadrature phase as discussed in [7, 9].
3
4. The polarity p varies as the scale changes; its behavior spatial averaging using a Gaussian at the selected scale.
in typical image regions can be summarized as follows: The three texture components are ac; pc, and c, computed
at the selected scale; the anisotropy and polarity are each
Edge: The presence of an edge is signaled by p holding modulated by the contrast in analogy to the construction
values close to 1 for all . of the color-cone coordinates. (Recall that anisotropy and
Texture: In regions with 2-D texture or 1-D flow, p decays polarity are meaningless in regions of low contrast.) In
with due to the presence of multiple orientations. effect, a given textured patch in an image first has its texture
properties extracted and is then replaced by a smooth patch
Uniform: In a constant-intensity neighborhood, p takes of averaged color. In this manner, the color and texture
on arbitrary values since the gradient vectors have properties in a given region are decoupled; for example, a
negligible magnitudes and therefore arbitrary angles. zebra is a gray horse plus stripes.
Note that in this formulation of the color/texture descrip-
The process of selecting a scale is based on the derivative tor, orientation and selected scale do not appear in the feature
of the polarity with respect to scale. First, we compute the vector; as a result, grouping can occur across variations in
polarity at every pixel in the image for k = k=2; k = scale and orientation.
0; 1; : : : ; 7, thus producing a “stack” of polarity images
across scale. Then, for each k, the polarity image computed
2.2 Grouping with the EM Algorithm
at scale k is convolved with a Gaussian with standard de-
viation 2 k to yield a smoothed polarity image p k . For
˜ Once an image has been processed using the above fea-
each pixel, we select the scale as the first value of k for ture extraction scheme, the result is a large set of 6-D feature
which the difference between successive values of polarity vectors, which we may regard as points in a 6-D feature
(p k , p k,1 ) is less than 2%. In this manner, we are per-
˜ ˜ space. In order to divide these points into groups, we make
forming a soft version of local spatial frequency estimation, use of the Expectation-Maximization (EM) algorithm [3] to
since the smoothed polarity tends to stabilize once the scale determine the maximum likelihood parameters of a mixture
window encompasses one approximate period. Since we of K Gaussians inside the 6-D feature space.
stop at k = 3:5, the largest period we can detect is approx- The EM algorithm is used for finding maximum likeli-
imately 10 pixels. Note that when the period is undefined, hood parameter estimates when there is missing or incom-
as is the case in uniform regions, the selected scale is not plete data. In our case, the missing data is the region to
meaningful and is set to zero. We declare a pixel to be which the points in the feature space belong. We estimate
uniform if its mean contrast across scale is less than 0:1. values to fill in for the incomplete data (the “E-Step”), com-
Another method of scale selection that has been proposed pute the maximum likelihood parameter estimates using this
[8] is based on localizing extrema across scale of an invariant data (the “M-Step”), and repeat until a suitable stopping cri-
of M , such as the trace or determinant. In this algorithm, terion is reached.
which is applied to the problem of estimating the slant and The first step in applying the EM algorithm is to initialize
tilt of surfaces with tangential texture, it is necessary to a mean vector and covariance matrix to represent each of
perform natural smoothing at a scale tied to the artificial the K groups. We initialize the means to random values
scale. We found that this extra smoothing compromised the and the covariances to identity matrices. (In earlier work
spatial localization ability of our scale selection method. we chose the initialization for EM carefully, but we have
found that the initialization has little effect on the quality
Texture features of the resulting segmentation.) The update scheme allows
Once a scale is selected for each pixel, that pixel is
for full covariance matrices; variants include restricting the
covariance to be diagonal or a constant times the identity
assigned three texture descriptors. The first is the polarity,
matrix. Full covariance matrices suit our problem, since
p . The other two, which are taken from M , are the
anisotropy, defined as a = 1 , 2 =1 , and the normalized
many plausible feature clusters require extruded covariance
p shapes, e.g. the shades of gray along the color cone axis.
texture contrast, defined as c = 2 1 + 2 .3 These are
Upon convergence, the Gaussian mixture parameters can
related to derived quantities reported in [8].
be inspected to determine what color/texture properties are
represented by each component of the mixture. Some ex-
2.1.3 Combining color and texture features amples of groups that can form include the following:
The color/texture descriptor for a given pixel consists of
six values: three for color and three for texture. The three bright, bluish, and textureless regions (e.g., sky)
color components are the color-cone coordinates found after
anisotropic and non-polar regions (e.g., zebra hide)
3 If we use a centered first difference kernel in the gradient computation,
the factor of 2 makes c range from 0 to 1. green weak-isotropic texture (e.g., grass)
4
5. We have thus far not discussed how to choose K , the
(a) (b)
number of mixture components. Ideally we would like to
choose that value of K that best suits the natural number of
groups present in the image. One readily available notion of
goodness of fit is the log-likelihood. Given this indicator, we
can apply the Minimum Description Length (MDL) princi- (c)
ple [22] to select among values of K . As a consequence of
this principle, when models using two values of K fit the
data equally well, the simpler model will be chosen. For our
experiments, K ranges from 2 to 5.
Once a model is selected, the next step is to perform
spatial grouping of those pixels belonging to the same
color/texture cluster. We first produce a K -level image
which encodes pixel-cluster memberships by replacing each (d)
pixel with the label of the cluster for which it attains the high-
est likelihood (see Fig. 2(d)). To enforce a small amount of
spatial smoothness in this representation, we apply a 3 3
maximum-vote filter to the raw cluster-membership image.
Finally, we run the resulting image through a connected- (e) (f)
components algorithm to produce a set of labeled image
regions (see Fig. 2(e)). (Alternatively, one could enforce
spatial constraints by appending the pixel coordinates to the
feature vectors, though we observed that this method too
often yields unsatisfactory segmentations.)
Figure 2. Creating the blobworld representation.
2.3 Describing the regions a Original image.
We store a simple description of each region’s color, b Scale estimated using polarity. The values range
texture, and spatial characteristics. (See Fig. 2(f) for a visu- from = 0 black to = 3:5 white.
alization of the stored representation.) c The six components of the color texture fea-
ture vectors, each bounded between 0 white and
Color and texture descriptors 1 black. Top: the locally smoothed color-cone co-
The two dominant colors within a connected component ordinates. Bottom: the texture coordinates; from
are chosen by using the EM algorithm to fit a mixture of
left to right, ac, pc, and c. The zebra hide is highly
two Gaussians in the HSV cone. The details are as before
anisotropic and in general has high texture contrast.
except that in this case we restrict the covariances to be a The polarity is largest around the edges, where the
constant times the identity matrix. Upon convergence, the shading gradient points primarily in one direction.
two mean vectors are recorded as the dominant colors in the d The results of clustering these feature vectors
region. When the color distribution inside the HSV cone
into K = 2; 3; 4; 5 groups using EM to learn a mix-
is in fact unimodal, both means become nearly coincident;
ture of Gaussians. Pixel cluster memberships are
we have not found it necessary to apply model selection shown as one of up to ve gray levels. The MDL
between K = 1 and K = 2. principle suggests that the rightmost image K = 5
For each image region (blob) we store the mean texture provides the best segmentation of the data. Most
descriptors (i.e., anisotropy, orientation, contrast) and the noticeable in this segmentation are oriented texture,
top two colors. We do not store the selected scale, since we
which is found throughout the zebra hide, and uni-
want to be invariant to scales in the range k = 0; : : : ; 3:5.
form or low-contrast texture, which accounts for
Although polarity is used for scale selection, we discard it most of the background.
e The segmentation for K = 5 after application of
a 3 3 max-vote lter. Each connected component
here, since in any textured or uniform region it is approxi-
mately zero by virtue of the scale selection process.
in this image which possesses an area greater than
2 of the total image area produces a blob.
f The blobworld representation. Each blob en-
codes summary information about the underlying
color, texture and shape properties.
5
6. Spatial descriptors or disjunction of compound queries (“like-blob-1 and like-
The geometric descriptors of the blob are simply the blob-2”). In the future, we might expand this definition to
centroid c and scatter matrix S of the blob region; the cen- include negation (“not-like-blob-1”) and to allow the user to
troid provides a notion of position, while the scatter matrix specify two blobs with a particular spatial relationship as an
provides an elementary shape description. In the querying atomic query (“like-blob-1-left-of-blob-2”).
process discussed in Section 3.1, centroid separations are Once a compound query is specified, we score each
expressed using Euclidean distance. The determination of database image based on how closely it satisfies the com-
pound query. The score i for each atomic query (like-blob-
p
the distance between scatter matrices is based on the three
quantities detS 1=2 = 1 2 , 1 , 2 =1 , and
7. . (1 i) is calculated as follows:
and 2 are the eigenvalues of S ;
8. is the argument of the
1. Find the feature vector vi for the desired blob bi . This
principal eigenvector of S .) These three quantities represent
vector consists of the stored color, texture, position,
approximate area, eccentricity, and orientation.
and shape descriptors.
3 Image retrieval by querying 2. For each blob bj in the database image:
Anyone who has used a search engine, text-based or oth- (a) Find the feature vector vj for bj .
erwise, is familiar with the reality of unwanted matches. (b) Find the Mahalanobis distance between vi
Often in the case of text searches this results from the use and vj using the diagonal covariance ma-
of ambiguous keywords, such as “bank” or “interest” [27]. trix (feature weights) set by the user:
1
, vj T Σ,1vi , vj
Unfortunately, with image queries it is not always so clear
why things go wrong. Unlike with text searches, in which dij = vi 2
.
the user can see the features (words) in a document, none (c) Measure the similarity between bi and bj using
dij
of the current content-based image retrieval systems allows ij = e, 2 . This score is 1 if the blobs are
the user to see exactly what the system is looking for in identical in all relevant features; it decreases as
response to a similarity-based query. Simply allowing the the match becomes less perfect.
user to submit an arbitrary image (or sketch) and set some
abstract knobs without knowing how they relate to the in- 3. Take i = maxj ij .
put image in particular implies a degree of complexity that
searching algorithms do not have. As a result, a query for The compound query score for the database image is cal-
a bear can return just about any object under the sun if the culated using fuzzy-logic operations [13]. For example, if
query is not based on image regions, the segmentation rou- the query is “like-blob-1 and (like-blob-2 or like-blob-3),”
tine fails to “find” the bear in the submitted image, or the the overall score for the image is minf1 ; maxf2 ; 3 gg.
submitted image contains other distinctive objects. Without The user can also specify a weighting i for each atomic
realizing that the input image was not properly processed, query. If “like-blob-i” is part of a disjunction in the com-
the user can only wonder what went wrong. In order to help pound query, the weighted score for atomic query i is
the user formulate effective queries and understand their re- 0 = i i ; if it is in a conjunction, its weighted score is
i
sults, as well as to minimize disappointment due to overly 0 = 1 , i 1 , i .
i
optimistic expectations of the system, the system should We then rank the images according to overall score and
display its representation of the submitted image and the return the best matches, indicating for each image which
returned images. set of blobs provides the highest score; this information
will help the user refine the query. After reviewing the
3.1 Querying in blobworld query results, the user may change the weighting of the blob
features or may specify new blobs to match.
In our system, the user composes a query by submitting an
image and seeing its blobworld representation, selecting the
3.2 Results
blobs to match, and finally specifying the relative importance
of the blob features. The user may also submit blobs from We have performed a variety of queries using a set of
several different images. (For example, a query might be 2000 images from the commercial Corel stock photo col-
the disjunction of the blobs corresponding to airplanes in lection. We used the following categories: African Spe-
several images, in order to provide a query that looks for cialty Animals; Air Shows; Arabian Horses; Bald Eagles;
airplanes of several shades.) Bears; Canadian Rockies; Caribbean; Cheetahs, Leopards,
We define an “atomic query” as one which specifies a Jaguars; China; Death Valley; Deserts; Elephants; Fields;
particular blob to match (e.g., “like-blob-1”). A “compound France; Kenya; Night Scenes; Sheep; Sunsets; Tigers; and
query” is defined as either an atomic query or a conjunction Wild Animals. Sample queries are shown in Figs. 3–4.
6
9. Figure 5. Tiger query performance.
Figure 3. Blobworld query for tiger images. 28 of
the top 50 images are tigers; tiger images make up
5 of the database. Figure 6. Zebra query performance.
Figure 7. Airplane query performance.
Figure 4. Blobworld query for zebra images. 24 of
the top 50 images are zebras, while less than 2 of
the images in the database are zebras. Figure 8. Sunset query performance.
7
10. 3.2.1 Comparison to color histograms [8] J. G˚ rding and T. Lindeberg. Direct computation of shape
a
We have compared our results to those obtained using cues using scale-adapted spatial derivative operators. Int. J.
color histogram matching, following the procedure of Swain Comp. Vis., 17(2):163–191, Feb 1996.
[9] G. H. Granlund and H. Knutsson. Signal Processing for
and Ballard [24]. The color histogram for each image uses
Computer Vision. Kluwer Academic Publishers, 1995.
8 divisions for the intensity axis and 16 for each oppo- [10] A. Gupta and R. Jain. Visual information retrieval. Comm.
nent color axis. Given a query image with histogram Qi , Assoc. Comp. Mach., 40(5):70–79, May 1997.
each database image (with histogram Di ) receives score
P [11] T. Hofmann, J. Puzicha, and J. M. Buhmann. Deterministic
i jQi , Di j. As before, we rank the database images and annealing for unsupervised texture segmentation. In Proc.
return the best matches. Figures 5–8 show how the precision Int. Workshop on Energy Min. Methods in Comp. Vis. and
changes as more images are returned; the blobworld query Patt. Rec., pages 213–228, 1997.
[12] A. K. Jain and F. Farrokhnia. Unsupervised texture segmen-
results are better than the color histogram results, except for
tation using Gabor filters. Pattern Recognition, 24(12):1167–
the tiger query. We believe the good color histogram results
1186, 1991.
for the tiger query occur largely because of the limited nature [13] J.-S. Jang, C.-T. Sun, and E. Mizutani. Neuro-Fuzzy and Soft
of the test database; few non-tiger images in this collection Computing. Prentice Hall, 1997.
have significant amounts of both orange and green. Adding [14] P. Kelly, M. Cannon, and D. Hush. Query by image example:
pictures of, say, orange flowers in a field would degrade the the CANDID approach. In SPIE Proc. Storage and Retrieval
color histogram performance without significantly affecting for Image and Video Databases, pages 238–248, 1995.
the blobworld performance. [15] T. Leung and J. Malik. Detecting, localizing and grouping
repeated scene elements from an image. In Proc. Eur. Conf.
Comp. Vis., pages 546–555, 1996.
4 Conclusions [16] P. Lipson, E. Grimson, and P. Sinha. Configuration based
We have proposed a new method which uses Expectation- scene classification and image indexing. In Proc. IEEE
Maximization on color and texture jointly to provide an Comp. Soc. Conf. Comp. Vis. and Pattern Recogn., pages
image segmentation, as well as a new image representation 1007–1013, 1997.
[17] J. Malik and P. Perona. Preattentive texture discrimination
(blobworld) which uses this segmentation and its associated
with early vision mechanisms. J. Opt. Soc. Am. A, 7(5):923–
descriptors to represent image regions explicitly. We have 932, 1990.
demonstrated a query mechanism that uses blobworld to [18] W. Niblack et al. The QBIC project: querying images by
retrieve images and help guide user queries. content using colour, texture and shape. In SPIE Proc. Stor-
age and Retrieval for Image and Video Databases, pages
Acknowledgments 173–187, 1993.
[19] V. Ogle and M. Stonebraker. Chabot: Retrieval from a re-
We would like to thank David Forsyth, Joe Hellerstein, lational database of images. IEEE Computer, 28(9):40–48,
Ginger Ogle, and Robert Wilensky for useful discussions Sep 1995.
related to this work. This work was supported by an NSF [20] A. Pentland, R. Picard, and S. Sclaroff. Photobook: Content-
Digital Library Grant (IRI 94-11334) and NSF graduate based manipulation of image databases. Int. J. Comp. Vis.,
fellowships for Serge Belongie and Chad Carson. 18(3):233–254, 1996.
[21] J. Ponce, A. Zisserman, and M. Hebert. Object Represen-
tation in Computer Vision—II. Number 1144 in LNCS.
References Springer, 1996.
[1] S. Ayer and H. Sawhney. Layered representation of motion [22] J. Rissanen. Modeling by shortest data description. Auto-
video using robust maximum-likelihood estimation of mix- matica, 14:465–471, 1978.
ture models and MDL encoding. In Proc. Int. Conf. Comp. [23] U. Shaft and R. Ramakrishnan. Data modeling and querying
Vis., pages 777–784, 1995. in the PIQ image DBMS. IEEE Data Engineering Bulletin,
[2] J. Big¨ n. Local symmetry features in image processing. PhD
u 19(4):28–36, Dec 1996.
thesis, Link¨ ping University, 1988.
o [24] M. Swain and D. Ballard. Color indexing. Int. J. Comp. Vis.,
[3] A. Dempster, N. Laird, and D. Rubin. Maximum likeli- 7(1):11–32, 1991.
hood from incomplete data via the EM algorithm. J. Royal [25] Y. Weiss and E. Adelson. A unified mixture framework for
Statistical Soc., Ser. B, 39(1):1–38, 1977. motion segmentation: Incorporating spatial coherence and
[4] P. Enser. Query analysis in a visual information retrieval estimating the number of models. In Proc. IEEE Comp. Soc.
context. J. Doc. and Text Management, 1(1):25–52, 1993. Conf. Comp. Vis. and Pattern Recogn., pages 321–326, 1996.
[5] W. F¨ rstner. A framework for low level feature extraction.
o [26] W. Wells, R. Kikinis, W. Grimson, and F. Jolesz. Adaptive
In Proc. Eur. Conf. Comp. Vis., pages 383–394, 1994. segmentation of MRI data. In Int. Conf. on Comp. Vis.,
[6] D. Forsyth, J. Malik, and R. Wilensky. Searching for digital Virtual Reality and Robotics in Medicine,pages 59–69,1995.
pictures. Scientific American, 276(6):72–77, June 1997. [27] D. Yarowsky. Word-sense disambiguation using statistical
[7] W. T. Freeman and E. H. Adelson. The design and use of models of Roget’s categories trained on large corpora. In
steerable filters. IEEE Trans. Pattern Analysis and Machine Proc. Int. Conf. Comp. Linguistics, pages 454–460, 1992.
Intelligence, 13(9):891–906, 1991.
8