Life Science Journal 2013;10(4s) http://www.lifesciencesite.com http://www.lifesciencesite.com lifesciencej@gmail.com 593 Content Based Image Retrieval by Shape, Color and Relevance Feedback Mussarat Yasmin, Sajjad Mohsin, Isma Irum, Muhammad Sharif COMSATS Institute of Information Technology, Islamabad, 44000, Pakistan mussaratyasmin@comsats.edu.pk Abstract: Efficient content based image retrieval has been proposed in this study by combining shape and color features and relevance feedback. In this era of digital communication, images are everywhere and these images consist of shape and color. For true image representation it is necessary to represent the shape correctly semantically. Only in this case accurate matching and retrieval can be performed. In these days navigation through image databases is very common. For correct image search and retrieval, the proposed method has been proved to be efficient and having better performance with the help of experimental results. Proposed method has also been compared with existing state of art methods that clearly shows its outperformance. [Mussarat Yasmin, Sajjad Mohsin, Isma Irum, Muhammad Sharif. Content Based Image Retrieval by Shape, Color and Relevance Feedback. Life Sci J 2013;10(4s):593-598] (ISSN:1097-8135). http://www.lifesciencesite.com . 91 Keywords: Color Histogram, Image Matching, Image Search, Probabilistic Weighting, Shape Descriptor 1. Introduction Image has achieved a very important place as a source of information sharing in this age of digital communication. “A picture is worth a thousand words“, directs a human mind towards interpreting the numerous and complex ideas through a single image. In this context the thing that matters is the perception i.e., how two people perceive the meaning or content described in the image. It is termed as semantic of image. Millions of digital images exist in the form of image databases or repositories in the world. Search is one of essential and key operations of database systems. Similarly anyone may need a specific image at any time according to the need or idea from these image repositories. As true perception is very important for image presented as source of information sharing, similarly true interpretation of image information is necessary for image matching and search. For the sake of these true interpretations of images, content based image retrieval (CBIR) systems have been devised through which the images are interpreted via the contents presented in the image. These contents are best possibly extracted in the form of digital attributes known as features. Some possible ways of extracting features from the image are color, shape, texture, order statistics etc. In this study a new state of art method has been built based on a new shape features extraction method, color histogram and relevance feedback. These three features have been combined to get advantage of their individual powerful characteristics at the same time and it comes up as a strong and efficient image description method for image matching and retrieval. 2. Existing Work A number of methods have been existed for image retrieval in literature as color descriptor, shape descriptor and methods with relevance feedback. Color descriptors use the color intensities of image existed in the form of RGB or in converted color spaces like HSV, HSL, and CMYK etc. Three approaches are used in extracting color information of images; first one considers the color of images globally, second one divides the image into partitions and takes the color information from individual segments and the third one divides the image into segments via segmentation algorithm [1]. Color histograms have been used for image matching [2]. Traditional color histograms use the binning methods and weight of each bin refers to the number of pixels falling into that bin. These methods do not take into account color distributions. In [3] an adaptive approach has been proposed to address the color distributions. Mathematical morphology based color descriptions have been defined in [4]. Local binary patterns combined with color histogram have been implemented for fast image matching and retrieval in [5]. Shape descriptors take the shapes presented in the image into account and convert the boundary or region information of these shapes into features. Fourier descriptors (FD) have been implemented in many applications of shape description due to their prominent features of simple derivation, normalization and noise robustness [6]. The curvature scale space descriptor (CSS) converts shape boundary into a 1D signal and performs analysis of this signal in scale space [7]. In [8] scale normalization has been performed on complex vector obtained from shape boundary. MPEG-7 adopted two