Image Retrieval Based on Color Saliency Histogram

Abstract

A new image retrieval method based on Color Saliency Histogram is proposed. Two features of a given image, the radius entropy and angle entropy, are defined and used to describe the spatial distribution of color, then, combining the entropies with the color coherence value, the saliency of each kind of color is calculated. After that, the Color Saliency Histogram is built for image retrieval. Experimental results show that the proposed method is feasible and more effective. Keywords-Color Saliency Histogram; spatial distribution entropy; radius entropy; angle entropy; Content Based Image Retrieval entropy (7), an Object-of-Interest may be segmented into different sections, and thus may be set by diverse weights. Then it may lead to unexpected retrieval result. To present more preferably the spatial distribution of color blocks in an image, this paper proposes a different way for spatial segmentation, namely, the color-block segmentation, where similar color pixels that are all connected one by one are marked as a unified block instead of separating them artificially by sectors or circular belt areas. Additionally, we define and calculate radius entropy and angle entropy of each kind of color, and then integrate the entropies with the color coherence values to represent color saliency. Thereafter, color saliency histogram is formed for image retrieval. Compared with the traditional color histogram and the method used in (8), experimental results show that our scheme is feasible and more effective.