Visual saliency detection based on color contrast and distribution

Abstract

This paper is concerned with color contrast and distribution for detecting salient regions. First, in order to improve the computational efficiency and reduce the disturbance of noise, the input image is pre-segmented into superpixels. Next, color contrast features are considered in Lab color space and opponency color space. The color distances between a superpixel and other superpixels are calculated, but we do not choose all superpixels to participate the difference. In the meanwhile, the distribution feature is shown by considering the rarity and position of pixels. Finally, we select 2D entropy to measure the performance of salient maps, and select the proper features to fuse. Experimental results show that the proposed method outperforms the state-of-the-art methods on salient region detection.