DCT-SVD domain feature vector for image retrieval

The novel approach combines Cosine Transform (DCT) and Singular Value Decomposition (SVD) for content based image retrieval (CBIR). DCT coefficients are mapped into four, eight, sixteen, thirty two and sixty four quadrants and then SVD is applied on each quadrant. The singular values from each quadrant are used as a feature vector for each image. Further image is divided into blocks and DCT applied on each block. Each block DCT coefficients are mapped into different quadrants and then SVD apply on each block. These SVD coefficients are used as a feature vector for each image in the database. Proposed algorithm tested over database of 1200 images having 15 different categories. Results are compared using grayscale image, RGB color plane and YCbCr color plane. Two similarity measures are used Bray Curtis Distance (BCD) and Euclidean Distance(ED). Performance evaluation of proposed method calculated by using overall average precision and overall average recall.