CONTENT BASED IMAGE RETRIEVAL USING DIRECTIONAL COLOR CORRELOGRAMS DEBASHIS DEBNATH Masters Research Scholar School of Education Technology Jadavpur University, Kolkata 700032, India devdebashis@sify.com Dr. RANJAN PAREKH Asst. Professor School of Education Technology Jadavpur University, Kolkata 700032, India rparekh@school.jdvu.ac.in Abstract: This paper proposes an automated system for content based retrieval of images based on color information. Color information is modeled using a color correlogram which defines the probability of a color i occurring in the neighborhood of another color j at a distance k from it. This paper further extends the concept to a directional correlogram by modeling the spatial color distribution along four different directions viz. horizontal, vertical, right-diagonal and left-diagonal compared to only a single direction as in standard correlograms. To reduce the dimensionality of the correlogram, color is quantized to six levels before creating the correlogram. Experimentations done using 120 images demonstrate the superiority of the proposed approach vis-à-vis the existing approaches. Keywords: Directional color correlogram; color based image retrieval; color quantization. 1. Introduction The last decade has seen a rapid growth in the use of multimedia components in various fields. Application areas like video on demand, computer based training materials, home entertainment, on-line gaming, digital art galleries etc. have led to the building up of large digital image repositories all around the world. In this scenario content-based search and retrieval techniques assume fundamental importance. An image database without an efficient retrieval mechanism is comparable to a library of books without a catalogue. Even though the information is present it is practically inaccessible to someone with specific search criteria. This paper outlines a methodology of building a content based image storage and retrieval system by automatically extracting and querying color related information from images. Color is a reliable and robust feature that can be effectively used for image indexing. It serves as a reliable feature even in presence of varying size and orientations of image. The most common method for color based recognition is the color histogram. However the main drawback of color histogram is that it only gives an overall indication of the color content of an image but does not provide information about the spatial distribution of the colors. Hence two different images with similar color distribution can produce similar histograms, again similar images with slight variations in contrast, brightness and color shades can create different histograms. 2. Previous Work Initial approaches of color based retrieval involved comparing RGB values of every corresponding pixel of two images [4]. This approach apart from being computation intensive was very sensitive to slight changes in lighting and camera position. In [6] image comparison and retrieval is done by comparing global histogram sets of two images. To capture local color information a local histogram approach is followed in [2] whereby the image is partitioned into a number of rectangular cells and local color histograms are computed for each cell. An alternative approach of color-signatures is presented in [1]: given a color if a cell is represented by it, the cell is assigned a value of 1 otherwise it is set to 0. This generates a bit-stream called a color-signature. Similarity of images is determined by intersection of their color signatures. Based on the observation that humans are prone Debashis Debnath et al. / International Journal of Engineering Science and Technology (IJEST) ISSN : 0975-5462 Vol. 3 No. 6 June 2011 4908