Image Retrieval Using Modified Color Variation Co-occurrence Matrix

Abstract

Texture is widely used as an important feature for content based image retrieval (CBIR). In this paper, color variation co-occurrence matrix (CVCM) modified from a previous investigation has been proposed for describing texture characteristics of an image. An image is processed and converted into four CVCMs based on overlapping 3 x 3 windows scanning from top to bottom and left to right. Each 3 x 3 window is divided into 4 grids with the central pixel located at four different corners. First, the original image with a size of N _x x N _y is converted into 4 motif matrices from individual scanning windows according to the traversal between the differences of four adjacent pixels. By contrasting to a previous method that 6 motif patterns were discriminated, an additional pattern was used to resolve ambiguity amounting to 7 motif patterns in total. By computing the probability of adjacent pattern pairs, an image retrieval system has been developed. The performance of the CVCM system was evaluated in the experiments using four different image sets. The experimental results reveal that the proposed CVCM-based image retrieval system outperforms the methods proposed by Huang and Dai and Jhanwar et al.