Saliency detection via boundary and center priors

This paper proposes a novel method for saliency detection in HSL color space. We firstly divide the traditional global contrast based saliency into background saliency and foreground saliency. Background saliency is then simplified into boundary saliency according to background priors. Furthermore, upper and lower boundary saliency maps are obtained. To overcome the defects of foreground saliency, we redefine it in this paper. Besides, an efficient method is proposed to pick up the center region to obtain center saliency and use it to represent foreground saliency. Finally, a well-founded fusion is proceeded to get the saliency map. Extensive experiments on MSRA dataset show that our algorithm has lower computation cost and higher performance compared with the other 10 state-of-the-art saliency detection methods.