17.xps

World Applied Sciences Journal 19 (3): 404-412, 2012 Content Based Image Retrieval: Survey
Mehwish Rehman, Muhammad Iqbal, Muhammad Sharif and Mudassar Raza COMSATS Institute of Information Technology, Wah Cantt, Pakistan Abstract: The requirement for development of CBIR is enhanced due to tremendous growth in volume of images
as well as the widespread application in multiple fields. Texture, color, shape and spatial layout are the underlying traits to represent and index the images. These peculiar features of images are extracted and implemented for a similarity check among images. The problem of content based image retrieval is based on generation of peculiar query. For relevant images that meet their information need, an automated search is initiated by drawing a sketch or with the submission of image having similar features. Similarity between extracted features can be measured by using different algorithms. The use of relevance feedback as a post retrieval step enhances the optimization of the process. The necessity to explore the ever growing volume of image and video is motivating the development of efficient CBIR algorithms. Different algorithms and models for the retrieval of images have been explored over the last twenty years. In this paper an analysis of visual contents of image is done with respect to features related to low level after extracting from image that are color, texture and shape. Here most popular algorithms of feature extraction and relevance feedback that try to bridge extracted low level features and features with high level semantics gap from image are discussed. A brief overview of an algorithm has been presented which is based on fuzzy logic and is used for the selection of Key words: CBIR Feature Extraction Relevance Feedback Semantic gap Sobel Edge Detection
INTRODUCTION
and geographic info system [5, 6], trademark databases, weather forecast [7], criminal investigations [8], image Technique based on image or visual contents usually classification [9, 11], image search over the Internet [12] referred as features for the purpose of searching images and remote sensing field for indexing biomedical images with respect to request and interest of user from large image databases. Since 1990s with the emergence and advancement of this field makes it possible to represent Definition: Visual features as color, shape and texture are
image by using low-level features instead of keywords. In implemented for retrieval of images. Traditional methods both theoretical research and system development of image indexing have been proven neither suitable nor remarkable progress has been made during past few years.
efficient in terms of space and time so it triggered the Still there are many unsolved problems in the area which development of the new technique. It is a 2 step process continue to attract the attention of researchers from where image features are extracted in first step to a various fields [1]. For various applications the deployment distinguishable extent. In second step matching of of huge databases has now become realisable as power of features which are visually similar is done.
processors increases and memories become cheaper. Professional areas like architecture, geography, Semantic Gap: The two retrieval systems namely, content
medicine, publishing databases of art works and satellite and text based retrieval systems differ in the sense that are attracting more users to access and utilize images is the indispensable part of latter system is human need of time now. For CBIR technology few strong interaction. High level features as keywords, text applications could be identified as architecture design [2], description uses by humans to measure similarity and art & craft museums [3], archaeology [4], medical imaging image interpretation. On the other hand the low Corresponding Author: Mehwish Rehman, COMSATS Institute of Information Technology, Wah Cantt, Pakistan.
World Appl. Sci. J., 19 (3): 404-412, 2012 levelfeatures with semantics [17, 18] usually color, shape, texture extraction is done automatically using computer vision techniques. System proposed in [19] was designed to cope with audiovisual queries [86] combining general approach to any real valued similarity measure for embedding in current CBIR systems [20]. In order to eliminate the semantic gap CLUE methodology is presented to retrieve image clusters which are semantically coherent. In other CBIR systems top matched target images are displayed to users. After giving coefficients is obtained and to compare Euclidean image as query, target image collections are chosen near distance is used to compare texture features & wavelet or similar to query image. These target images can be clustered by using NCut clustering into different semantic classes by putting image of same semantic in one cluster.
Texture Feature Extraction: In CBIR, one of the salient
Then the image clusters is displayed by the system and common attribute is texture. In image classification [30, similarity measure model is adjusted with respect to feed 31], texture provides important information as in many images of real world. In highlighting high-level semantics, texture is salient feature for purpose of image retrieval.
Low Level Image Features: Milestone of CBIR system is
Systems comprises spectral features achieved by using low level feature extraction [22]. Feature Extraction can be Gabor filtering, wavelet transform texture features done from region or the entire image. Mostly, users are commonly used by system of retrieving images [32]. For concerned with particular region within the image rather retrieval of images Gabor and wavelet features are widely than the whole image. In general, CBIR algorithms are used among various other texture features. For region specific. Representation of image is similar to rectangular shaped images Gabor filter and wavelet system of human being perception at region level.
transform are basically designed. In CBIR system Retrieval based on global features is comparatively arbitrary-shapes regions exists [33]. In Tiled approach, the simpler. Region-based image retrieval is focused in this matching is done in corresponding manner and query paper. Firstly image segmentation implement then from image tiles are compared against tiles of target image.
these segmented regions color [23, 24], texture, shape Matching of query image is allowed to any target image [25, 26] or spatial location can be extracted. CBIR tile. Based on the textured image, an algorithm was proposed in [27] exploited two different levels of features proposed which modeled the textural information by a set that is global and local to achieve better accuracy.
of features possessing a perceptual intelligence to be used for image retrieval [34].In medical field Texture image Extraction of Color: In image retrieval, the color is
retrieval is usually more demanded in many applications.
commonly used feature. Varieties of color spaces are In analyzing texture as one of the multi-resolution filtering available and different colors are defined on selected color method 2-D Gabor Filter proved to be very useful and space. To make image recognition possible by human, important feature is color. It is a property that relies on light reflection to eyes and the information processing in Bayesian Framework Texture Image Retrieval: Mostly
the brain. Focusing on color description and studying the images consist of not only the visual ingredients but also potential of morphological operators for content some texture hence a significant part of CBIR is dedicated description, main properties of color from a descriptive for the categorization of texture. A critical aspect of any point of view have been determined and a state-of-the-art CBIR implementation is its processing efficiency. In ordering approach has been implemented for the traditional systems, the total processing time is mainly extension of mathematical morphology to color images dependant on the query process. Therefore, total time [28]. System based on color & texture features has been needed to accommodate a query is a linear function of the developed. For color features RGB color histogram is used total number of images there in the data bank. In [36], & to determine texture features statistical texture measures authors proposed a technique based on Bayesian CBIR & wavelet transform is used. In order to make algorithm which drastically reduced the computational comparisons between color histogram Bhattacharyya World Appl. Sci. J., 19 (3): 404-412, 2012 Perceptual Features for Texture Representation and
Low Level Visual Features: To map sample of high
Retrieval: The main drawback of majority of already
dimension to dimension of low space biased discriminant existing techniques, whether they are statistical, structural Euclidean embedding and its semi-supervised extension or hybrid is that they are not efficient in computational uses. Both inter and intra class geometry are preserved by cost. A very significant computation cost is demanded by BDEE. Various experiments proved that superior to the some of these methods contrary to this for almost all popular relevance feedback dimensionality reduction types of textures, human visual perception seems to work algorithms, biased discriminant analysis BDA directed and marginal biased analysis and support vector machine SVM [48] based relevance feedback algorithms which are Relevance Feedback (RF): Relevance feedback is a
RF dimensionality reduction algorithm [49]. IKSVM significantly important algorithm which attempts to reduce algorithm proposed in[50] shows that time complexity is the gap between the two levels of features, namely high reduced to O(nlogm) and space complexity is O(nm). and low. In [38], an online algorithm of feature selection in Finally they are successful in formulating an approximate the RF learning scheme is proposed by Jiang et al. In this method which has space and time complexity devoid of technique, the outcomes of a search are presented to support vectors amount that is O(n). Hence they proved users and they are permitted to choose related or different that classification using IKSVM is efficient in terms of objects. RF process scenario is depicted in [39]. Through resources required. Using technology of classification, query-by- example or sketch, system provided initial multiple feature distances are joined to get similarity of an retrieval results. According to query, user judges whether image. A new strategy of two steps which incorporated and how much this image is similar (positive) or dissimilar data cleaning and noise tolerant classifier for managing (negative) to query image. At last, to learn feedback, the noisy positive examples. For validity of proposed machine learning algorithm is applied. To make RF scheme efficiency extensive experiments carried out on technique more robust and to change the query, use of two different collection of real image [51]. Most of the negative objects gives more options.
systems developed so far to retrieve image solely depend Discriminant Expectation-Maximization algorithm is on only one peculiar feature for the extraction of related proposed and the issue of image retrieval is treated as a images, where as a suitable conjunction of related features semi-supervised learning.[40]. GM models [41] of the can possibly culminate in improved performance.
images based on a universal GM model in a Bayesian Requirement for computation and space in nonlinear manner and information is extracted from the entire kernels is linearly proportional to support vectors number database. As a distance measure between GM and an which also increases with increase in training data for SVM classifier, Fast KL approximation is used with an classification [52]. Visual features are formed from small appropriate kernel function employed in each RF round to chunks of visual information in most of content based perform the RF task [42]. Methods include the naive system by using two types of techniques, namely fixed- Bayesian technique used in [43]. Greedy EM [44] avoids block partition [53, 56] and image segmentation [57, 59].
the overcome of the strong dependence of the solution on parameter initialization by incrementally adding Image Segmentation: Automatic image segmentation
components to the mixture until the desirable number of methods have been developed such as curve evolution, components has been reached. To measure parameters of graph partitioning and energy diffusion [60]. For images ideal query many current systems use only the low-level containing only similar color regions in color space such image features without using image semantic contents.
as direct clustering many pre-existing segmentations System will give more accurate results if feature vector perform better. These retrieval systems are applicable for define query in a better way otherwise more relevant working with colors only. One image region can be several results are not provided. GM model use the standard EM times matched to region of another image by an integrated algorithm and information is extracted from the entire region matching. On the significance matrix similar to database [45]. Relevance feedback focuses on selection of EMD [61] defining overall similarity of two images with features which merges the probabilistic formulation by heterogeneous color and texture ranges of natural images.
using both the positive and negative example. Algorithm In defining high-level concepts, texture is main factor.
learns the necessity to assign features of images with the Estimation of parameter of texture model is more user interaction and results are also applied [46].
problematic task and is required by majority of texture Considerable improvement in the performance of the CBIR segmentation algorithms. These demerits are overcome by JSEG segmentation in which contrary to estimation World Appl. Sci. J., 19 (3): 404-412, 2012 pattern of given color and texture, similarity is tested.
relationship representation is 2D-string and its variants Firstly, quantization of image colors into classes then (left/right, above/below) between objects. Semantic pixels are substituted by their respective label of the class content representations of images and directional of color to achieve mapping of class. Then on this image relationships are not sufficient alone.
map, spatial segmentation is performed. As a result, region with similar color texture is obtained beneficial for Object Recognition for Image Retrieval: Main hurdle and
various systems [62]. Blobworld segmentation [63] is difficulty in computer vision field in various implications widely implemented algorithm of segmentation where like image annotation [69] is recognition of object in pixels are grouped in a joint color texture-position used to images [70, 71]. Supervised or unsupervised algorithms to obtain it. At first the joint model with combination of be used for recognizing objects for semantic based image Gaussians distribution, position features are constructed.
retrieval have been developed. To recognize and learn Secondly by expectation maximization algorithm, object class models, learning method of unsupervised parameters of model are measured. Segmentation of image scale-invariant is presented from unsegmented/unlabelled is provided by resulting cluster of pixels. For object cluttered scenes. All aspects of the shape of objects segmentation from images with connectivity constraint, depicted by probabilistic representation are modelled as k mean is an extension of k-means algorithm. For k-means flexible constellations. For image classification and algorithm new centroid is defined for each region recognition, this model is used in a Bayesian way.
participating in working. A method [64] Field Excellent performance over range of datasets Programmable Gate Arrays can be worked for optimality demonstrated that model is quiet flexible.
of retrieval problem which provide dedicated functional Generative/discriminative is a two phase learning blocks to perform complex image processing operations.
approach with feature of multiple types to be recognized.
Novel image segmentation algorithm namely, Fuzzy Edge Development of classification method for CBIR [72] is Detection and Segmentation FEDS is built into an FPGA main goal. Description length of images normalized in generative phrase contains arbitrary number of extracted features. Classifier learns in discriminative phase. Multiple Shape-based Image Retrieval: In retrieval systems, shape
types of abstract regions can be used for Image is useful image features. Indexing and their performance representation. Over its feature space, each abstract of retrieval require number of coefficients. For shape region is a structured combination of Gaussian representation Fourier, curvature scale space, angular distributions. Regions can be taken from various radial transform (ART) and image moment descriptors are segmentation processes used in recognition region, called discussed. Using standard methods, the four shape the abstract region. Location of objects in each image is descriptors are checked against each other and the two not needed to be known. The efficiency of approach is most appropriate and available databases [66]. After demonstrated on a set of 860 images.
the segmentation of image into regions or objects according to features. Shape features usage for image Wavelet-Based CBIR: Multidimensional wavelet filter
retrieval is limited to specific application as it is not bank is a method adopted for any specific problem. Non- generalized so it is not easy to get accurate error free separable lifting scheme framework is mile stone of segmentation [67]. Outer boundary of the shape is method being proposed. It permits the design of filter required in boundary shaped representation. Entire shape bank having required number of degrees of freedom [73].
of region is utilized by region-based shape representation.
To satisfy the bi orthogonality condition and to maximize the regularity of both the primal wavelet for Spatial Location: In region classification, spatial location
analysis and the dual wavelet (used for synthesis), is also important factor besides color and texture for wavelet filter coefficient are designed [74]. Signal with a example, two things sharing same textural and color family of basis functions is decomposed by wavelet features having different spatial location defined with transform with the help of dilation and translation of respect to region’s location in image as top, upper and a mother wavelet. Wavelet transforms of a 2D signal bottom. Information related to spatial location is provided can be computed by recursive filtering and sub-sampling.
by minimum bounding rectangle and centroid of region Low frequency is denoted by L and H denotes high [68]. To derive semantic features, absolute spatial location frequency provides a multi-resolution scheme which is is not needed rather relative spatial relationship is more achieved by wavelet transform to texture analysis and valuable. Structure commonly used for directional World Appl. Sci. J., 19 (3): 404-412, 2012 Query Language Design: In bridging the semantic gap
also be used to measure the accuracy of the algorithm [75], query mechanism played a vital role. Many problems [83]. Higher the number of crossover points better will be related to query scenario that was conventional like the performance of the system. Precision-scope curve query-by-example QBE and query-by-sketch QBS can be implemented by many developers is used to evaluate addressed by a specialized query language designed for image retrieval performance [84]. Rank measure is another CBIR. Using multiscale color coherent vector, the performance measure. Performance will be better if the semantic content is captured and texture features computation is done from decomposing wavelet. Query language is hard to understand and too much attention is Image Databases: Image Collection with various contents
needed in reducing and minimizing gap related to similar to natural sceneries presented in corel image database. Domain professionals pre-classified these In [76], querying databases, natural language of images into different categories. In order to test retrieval query is designed. Vocabulary query language depends performance subset of Corel image dataset is used by half on semantic indicator that is elementary semantic system. In perceptual texture feature Brodatz textures [80] category. It is simple and more expressive. About images is widely used. Image retrieval that is web based, where for constructing sentence expressing an assertion words images are collected from internet. Natural images are can be used. During retrieval process images in appreciated by many researchers for semantic extraction.
databases are tested and only assertions are chosen.
Main causes are that the types of objects are limited.
Secondly in analysing, shape features are not so much Web Based Image Retrieval: Image retrieval based on
important. Due to inaccuracy of segmentation this semantics is main benefit in web image retrieval. Clear weakness can be avoided from shape features in high- hierarchical structure with some image information such as category of image included in URL of image file. Image title in HTML document also shows some beneficial Comparison of Working Performance: Comparison of
information. Image can be annotated up to certain extent the performance of different CBIR systems is not easy by such information. Only on textual evidences, existing task due to query difference [85]. Performance of Sobel, Web searcher such as web Seer [77] Google, Alta Vista Canny and Prewitt gradient operators is improved for all search image data base [78]. Many relevant images can be transforms in all proposed CBIR methods compared.
found by these approaches. As it is not confirmed Individual gradient operators used for image transforms.
whether or not the resultant images fulfill the query Kekre Transform Perform better in Prewitt operator. Best requirements so retrieval precision is not good. Process is performance shown by Hartley and Slant transforms in time consuming due to entire list through which user have Sobel operator. For Robert operator DCT proves to be to go for the desired images. Multiple topics also found which are mixed together. Developers are working and struggling to improve performance of web image retrieval.
CONCLUSION
For web image retrieval Unified relevance feedback framework is discussed in [79]. In following three aspects This paper reviewed the main CBIR components this framework is better and suitable than traditional RF including image feature representation, indexing and mechanism. They used both textual and visual features in system design, while highlighting the past and current a sequential way during the RF process.
technical achievements. Many research issues are highlighted and directions for future solution to cope with Performance Evaluation: In CBIR system, retrieval
these problems are suggested. Efficient multidimensional performance can be measured by precision and recall.
techniques are required for the retrieval system to enable Ratio of the Nr to k is precision. Where (N ) them fast and scaleable. The development of speedy, required images retrieved. Total number of extracted inexpensive and strong processors coupled with fast images is K [81]. Recall is recovered related images Nr memory devices have contributed a lot in this field.
divided by the total related images. Both Pr and Re should Hence, immense range applications are guaranteed by this be high. Pr(Re) graph uses to represent performance development using CBIR in future. Maximum support is rather than using Pr or Re individually. In precision and also provided in bridging the ‘semantic gap’ between low recall, crossover is the point on the graph where both the level features and the perceptual knowledge present in the precision and recall curves meet. The crossover point can images with the richness of human semantics.
World Appl. Sci. J., 19 (3): 404-412, 2012 REFERENCES
12. Kekre, H.B. and Sudeep D. Thepade, 2009.
Improving the Performance of Image Retrieval using 1. Long, F., H. Zhang and D.D. Feng, 2003.
Partial Coefficients of Transformed Image, Fundamentals of content-based image retrieval, in International Journal of Information Retrieval, Serials Multimedia Information Retrieval and Management Technological Fundamentals and Applications. New 13. Comaniciu P. Meer, D. Foran and A. Medl, 1998.
Bimodal system for interactive indexing and Retrieval 2. Kekre, H.B. and Sudeep D. Thepade, 2008. Creating of pathology images, In Workshop on Applications the Color Panoramic View using Medley of Grayscale and Color Partial Images, WASET International 14. Sameer Antani, L. Rodney Long and George R.
Journal of Electrical Computer and System Thoma, 2004. Content- Based Image Retrieval for Large Biomedical Image Archives, MEDINFO 2004 3. Kekre, H.B. and Sudeep D. Thepade, 2008. Color M. Fiesehi et al. (Eds) Amsterdam: IOS Press, Traits Transfer to Gray scale Images, In Proc of IEEE First International Conference on Emerging Trends in 15. Kom, F., N. Sidiropoulos, C. Faloutsos, E. Siegel and Z. Protopapas, 1998. Fast and effective Retrieval of 4. Kekre, H.B. and Sudeep D. Thepade, 2008. Scaling medical tumor shapes, IEEE Trans. on Knowledge Invariant Fusion of Image Pieces in Panorama and Data Engineering, 10(6): 889-904.
Making and Novel Image Blending Technique, 16. Sinha, U., A. Ton, A. Yaghmai, R.K. Taira and International Journal on Imaging (IJI), 1(A08): 31-46.
H. Kangarloo, 2001. Image Content Extraction: 5. Müller, H., N. Michoux, D. Bandon and A.
Application to MR Images of the Brain, Radio Geissbuhler, 2004. A review of content-based image Retrieval systems in medical applications – clinical 17. Lars Jacob Hove, 2004. Extending Image Retrieval benefits and future directions, International Journal Systems with a Thesaurus for Shapes, Masters 6. Lehmann, T., B. Wein, J. Dahmen, J. Bredno, F.
Vogelsang and M. Kohnen, 2000. Content-Based 18. Wilson, C., B. Srinivasan and M. Indrawan, 2000.
Image Retrieval in Medical Applications: A Novel A General Inference Network Based Architecture Multi-Step Approach, International Society for For Multimedia Retrieval, In Proceedings of Optical Engineering, 3972(32): 312-320.
IEEE International Conference on Multimedia 7. Kekre, H.B., Sudeep D. Thepade and Akshay Maloo, pp: 347-350, New York City, USA, 2000.
2010. Performance Comparison for Face Recognition 19. Roland Kwitt, Peter Meerwald and Andreas Uhl, using PCA, DCT &Walsh Transform of Row Mean 2011. Efficient Texture Image Retrieval Using Copulas and Column Mean, ICGST International Journal on in a Bayesian Framework, IEEE transactions on image Graphics, Vision and Image Processing (GVIP), 10: II.
8. Jarrah, K., M. Kyan, S. Krishnan and L. Guan, 2006.
20. Yixin Chen, 2005. Member, IEEE, James Z. Wang, Computational Intelligence Techniques and Their Member, IEEE and Robert Krovetz, CLUE: Cluster- Applications in Content-Based Image Retrieval, IEEE Based Retrieval of Image by Unsupervised Learning, Int. Conf. on Multimedia & Expo, pp: 33-36, Toronto, IEEE transactions on image Processing, 14.
21. Long, F., H. Zhang and D.D. Feng, 2003.
9. Antani, S., R. Kasturi and R. Jain, 2002. A survey on Fundamentals of content-based image retrieval, in the use of pattern recognition methods for Multimedia Information Retrieval and Management Abstraction, indexing and retrieval of images and Technological Fundamentals and Applications. New video, Pattern Recognition, 35(4): 945-965.
10. Smith, J.R. and C. Li, 1999. Image classification and 22. Laaksonen, J., E. Oja, M. Koskela and S. Brandt, 2000.
querying using composite region templates, Analyzing low-level visual features using content- Computer Vis. Image Understand, 75: 165-174.
based image retrieval, In: Proc int’l conf neural 11. Vailaya, A., M.A. T. Figueiredo, A.K. Jain and information processing, Taejon, 14-18.
H.J. Zhang, 2001. Image classification for content- 23. Huang, J., R. Kumar, M. Mitra, W.J. Zhu and based indexing, IEEE Trans. Image Processing, R. Zabih, 1997. Image Indexing Using Color Correlograms. In Proceedings of CVPR.
World Appl. Sci. J., 19 (3): 404-412, 2012 24. Sebe, N. and M. S. Lew, 1999. Robust color indexing.
37. Amadasun, M. and R. King, 1989. Textural features In Proceedings of the 7th ACM International corresponding to textural properties, IEEE Trans.
25. Schmi, C. and R. Mohr, 1997. Local gray value 38. Huang, T.S., X.S. Zhou, M. Nakazato, Y. Wu and invariants for image retrieval. IEEE Transaction on I. Cohen, 2002. Learning in content-based image Pattern Analysis and Machine Intelligence, Retrieval, Proceedings of the 2nd International 26. Zhou, X.S., Y. Rui and T. Huang, 1999. Water-Fil ing: a novel way for image structural feature Extraction. In 39. Zhu, X.S. and T.S. Huang, 2003. Relevance feedback Proceedings of IEEE International Conference on in image retrieval: a comprehensive review, 27. Nandagopalan Dr. S., B.S. Adiga and N. Deepak, 40. Tzotsos, A. and D. Argialas, 2008. A support vector 2008. A Universal Model for Content-Based Image machine approach for object based image Analysis.
Retrieval, World Academy of Science, Engineering Support Vector Machine Classification for Object- Based Image Analysis, Springer-Verlag, Chapter, 7.2.
28. Erchan aptoula and Sébastien lefèvre, 2009.
41. Duan, K.B. and S.S. Keerthi, 2005. Which is the best Morphological description of color images for multiclass svm method? An empirical study, content-based image retrieval, IEEE transactions on 42. Ritendra datta, Dhiraj joshi, Jia li and James Z. Wang, 29. Bikesh Kr. Singh and Bidyut Mazumdar, 2010.
2008. Image Retrieval: Ideas, Influences and Trends Content Retrieval from X-RAY Images Using Color of the New Age, ACM Computing Surveys, Vol. 40, &Texture Features, International Journal of No. 2, Article 5, Publication date: April 2008.
Electronics Engineering, 2(1): 25-28.
43. Liu, B., Y. Dai, X. Li, W. Lee and P. Yu, 2003. Building 30. Huang, J., S.R. Kumar and R. Zabih, 1998. An text classifiers using positive and unlabeled automatic hierarchical image classification scheme, in Examples, in Proc. IEEE Int. Conf. Data Mining, Proc. ACM Int. Conf. Multimedia, pp: 219-228.
31. Smith, J.R. and C. Li, 1999. Image classification and 44. Vlassis, N. and A. Likas, 2002. A greedy EM algorithm for Gaussian mixture learning, Neural querying using composite region templates, Computer Vis. Image Understand. 75: 165-174.
45. Marakakis, A., G. Siolas, N. Galatsanos, A. Likas and 32. Wang, J.Z., J. Li and G. Wiederhold, 2001.
A. Stafylopatis, 2011. Relevance feedback approach SIMPLIcity: semantics-sensitive integrated matching For image retrieval combining support vector for Picture libraries. IEEE Trans. Pattern Anal. Mach.
machines and adapted Gaussian mixture models, IET 33. Vautrot, P., N. Bonnet and M. Herbin, 1996.
46. Mohammed Lamine Kherfi and Djemel Ziou, 2006.
Comparative study of different spatial/spatial- Relevance Feedback for CBIR: A New Approach requency Methods (Gabor filters, wavelets, wavelets Based on Probabilistic Feature Weighting With packets) for texture segmentation/classification, in Positive and Negative Examples, IEEE Transactions Proc. Int. Conf. Image Processing, 3: 145-148.
34. Noureddine Abbadeni, 2011. Computational 47. MacArthur, S.D., C.E. Brodley and C. Shyu, 2000.
Perceptual Features for Texture Representation and Relevance feedback decision trees in content Based Retrieval, IEEE transactions on image processing, 20.
image retrieval, In Proceedings of IEEE Workshop 35. Grigorescu, S.E., N. Petkov and P. Kruizinga, 2002.
Comparison of Texture Features Based on Gabor 48. Tao, D., X. Li and S. Maybank, 2007. Negative Filters, IEEE Transactions on Image Processing, samples analysis in relevance feedback,, IEEE Trans.
36. Roland Kwitt, Peter Meerwald and Andreas Uhl, 49. Wei Bian and Dacheng Tao, 2010. Member, IEEE 2011. Efficient Texture Image Retrieval Using Copulas Biased Discriminant Euclidean Embedding for in a Bayesian Framework, IEEE transactions on image Content-Based Image Retrieval, IEEE transactions on World Appl. Sci. J., 19 (3): 404-412, 2012 50. Maji, S., A.C. Berg and J. Malik, 2008. Classification 62. Jing, F., M. Li, L. Zhang, H.J. Zhang and B. Zhang, using intersection kernel support vector Machine is 2003. "Learning in region-based image retrieval", efficient, IEEE Computer Vision and Pattern Proceedings of the International Conference on 51. Jun Zhang and Lei Ye and Senior Member, 2009.
IEEE "Content Based Image Retrieval Using Unclean 63. Carson, C., M. Thomas, S. Belongie, J. Hellerstein Positive Examples IEEE transactions on Image and J. Malik, 1999. Blobworld: "A System for Region-Based Image Indexing and Retrieval", Third 52. Steinwart I., 2003. "Sparseness of support vector International Conference on Visual Information machines-some asymptotically sharp bounds" Systems (VISUAL '99), volume 1614, Amsterdam, 53. Monay, F. and D. Gatica-Perez, 2007. "Modeling 64. Sriramakrishnan, C. and A. Shanmugam 2011. "Image semantic aspects for cross media image indexing," Retrieval Optimization Using FPGA Based Fuzzy IEEE Trans. Pattern Anal. Machine Intelligent, Segmentation" European Journal of Scientific Research ISSN 1450-216X, 63(1): 56-62.
54. Carneiro, G., A.B. Chan, P.J. Moreno and N.
65. Nakano, K. and E. Takamichi, 2003. "An Image Vasconcelos, 2007. "Supervised learning of semantic Retrieval System Using FPGAs," Proc. Asia and Classes for image annotation and retrieval," IEEE South Pacific Design Automation Conference, Trans. Pattern Anal. Mach. Intell, 29(3): 394-410.
66. Amanatiadis, A., V.G. Kaburlasos, A. Gasteratos amd 55. Theoharatos, C., G. Economou, S. Fotopoulos and S.E. Papadakis, 2011. "Evaluation of shape N.A. Laskaris, 2005. "Color-based image retrieval descriptors for Shape-based image retrieval" IET Using vector quantization and multivariate graph matching," in Proc. IEEE Int. Conf. Image Processing, 67. Gary, J.E. and R. Mehrotra, 1662. "Shape similarity-based retrieval in image database 56. Daptardar, A.H. and J.A. Storer, 2006. "Reduced systems," Proc. Of SPIE, Image Storage and Retrieval complexity content-based image retrieval using Vector quantization," in Proceeding Data 68. Shirahatti, N.V. and K. Barnard, 2005. "Evaluating Image Retrieval", Proc. IEEE Int. Conf. on Computer 57. Pappas, T.N., J. Chen and D. Depalov, 2007.
Vision and Pattern Recognition, I: 955-961.
69. Allan Hanbury, 2008. "A survey of methods for segmentation and semantic classification," IEEE image annotation", Journal of Visual Languages and 58. Frigui, H. and J. Caudill, 2006. "Unsupervised image 70. Serre, T., L. Wolf and T. Poggio, 2005. "Object segmentation and annotation for content-based recognition with features inspired by visual cortex," Image retrieval," in Proc. IEEE Conference On Fuzzy 71. Mutch, J. and D.G. Lowe, 2006. "Multiclass 59. Banerjee, S., G. Harit and S. Chaudhury, 2007. "Word object recognition with sparse, localized features," in image based latent semantic indexing for Conceptual querying in document image databases," in Proc. 9th 72. Chen, Y., X. Zhou and T. Huang, 2001. "One-class Int. Conf. Document Analysis and Recognition, SVM for learning in image retrieval," in Proc. Int.
Conf. Image Processing, Thessaloniki, Greece, 60. Hoiem, D., R. Sukhtankar, H. Schneiderman and L.
Huston, 2004. "Object-Based Image retrieval Using 73. Gwénolé Quellec, Mathieu Lamard, Guy Cazuguel, Statistical structure of images", Proc CVPR, Member, IEEE, Béatrice Cochener and Christian 61. Rubner, Y., C. Tomasi and L.J. Guibas, 1998. "A Roux, Fellow, IEEE "Adaptive Nonseparable metric for distributions with applications to image Wavelet Transform via Lifting and its Application to Databases," in Proc. IEEE Int. Conf. Computer Content-Based Image Retrieval" IEEE transactions World Appl. Sci. J., 19 (3): 404-412, 2012 74. Kovacevic, J. and W. Sweldens, 2000. "Wavelet 82. Dr Kekre, H.B. and D. Mishra, 2010. "Content Based families of increasing order in arbitrary dimensions," Image Retrieval using Weighted Hamming Distance" IEEE Trans. Image Processing, 9(3): 480-496 Proceedings of International Conference on 75. Deserno, T.M., S. Antani and L.R. Long. 2008.
Contours of Computing Techonolgy Babasaheb "Ontology of Gaps in Content-Based Image Gawde Institute of Technology, Mumbai, India on Retrieval" J Digital Imaging. February 2008. DOI: 83. Dr Kekre, H.B. and D. Mishra, 2010. "Performance 76. Bimbo, A.D., et al., "Query by dialog: an iterative Comparison of Four, Eight & Twelve Walsh approach to pictorial querying," Image Vis.
Transform Sectors Feature Vectors for Image Retrieval from Image Databases" Vol 2(5), Published 77. Frankel, C., M. Swain and V. Athitsos, 1996.
in International Journal of Engineering Science and "Webseer: An image search Engine for the World Wide Web", technical report TR-96-14 1996.
84. Huijsmans, D.P. and N. Sebe, 2005. S"How to 78. Jansen, B., J. Spink and A. Pedersen, 2003."An complete performance graphs in content based analysis of multimedia searching on AltaVista". In image Retrieval: add generality and normalize scope" Proceedings of the ACM SIGMM International IEEE Transaction on pattern analysis and Machine Workshop on Multimedia Information Retrieval (MIR) at the International Multimedia Conference.
85. Brown, L. and L. Gruenwald, 1988. "Tree-based 79. En Cheng, Feng Jing and Lei Zhang 2009. "A Unified indexes for image data," Vis. Communication Image Relevance Feedback Framework for Web Image Retrieval" IEEE transactions on image Processing, 86. Naphade, M.R. and T.S. Huang, 2002. "Extracting semantics from audiovisual content: The final 80. Brodatz, P. Textures, 1996. "A Photographic Album Frontier in multimedia retrieval," IEEE Trans. Neural for Artists & Designers, Dover, New York, NY 1966.
81. Mueller, H., S. Marchand-Maillet and T. Pun, 2002.
"The truth about Corel-evaluation in imageretrieval", Proceedings of the InternationalConference on Image and Video Retrieval (ICIVR),pp: 38-49.

Source: http://www.2irg.asia.cm/upload/CBIRWASJ.pdf