A Novel Approach for Effective Image Retrieval using Color, Texture, Edges and Valleys in Covariance Descriptor

Abstract

This paper presents a novel system for image retrieval based on integrated mean and covariance matrices of color, texture, edges and valleys information which are estimated from R G and B components of the image, Block variation of local correlation coefficients (BVLC) and Block difference of inverse probabilities (BDIP) on R, G and B components of the image respectively. The BVLC estimates the texture smoothness based on differences among local correlation coefficients and BDIP estimated the edges and valleys by local intensity maxima and local intensity minima respectively. The proposed integrated color, texture, edges and valleys information is categorized by Radial Basis Function Neural Network (RBFNN) to decrease the search space and to increase accuracy of retrieval. Chernoff measure is adopted to estimate the degree of divergence between query and target images. The proposed system has been estimated on Corel- 1k, 5k and 10k databases attained not only significantly better results but also the reasonable computation cost.