Fusing region contrast and graph regularization for saliency detection

Automatic detection of salient object from a static image is a highly active area of computer vision research. In this paper, we propose an effective region-contrast based solution for saliency estimation which involves three phases. First, we abstract an image into perceptually homogeneous regions to better capture structural information of the input image. Next, three kinds of region contrast measures, i.e., global distinctness, region compactness, and center prior, are evaluated and integrated together by means of a two-layer saliency structure to generate the initial saliency value of each image region. Lastly, we utilize a graph-based regularization algorithm to refine the initial saliency map and to encourage continuous saliency values across similar image regions, thus yielding a perceptually consistent saliency map. Extensive experiments on two publicly available benchmark databases demonstrate the advantage of the proposed method against fourteen state-of-the-art approaches in terms of detection accuracy and computational efficiency.