Salient object detection using contrast and background priors

Existing approaches are inefficient for detecting saliency maps of complex pictures. To address this problem, we propose a salient object detection algorithm using contrast and background priors. Firstly, the source image is segmented into perceptually uniform patches. Then we define contrast priors as salient edge, patches' global contrast and spatial distribution of patches with similar colors. Background prior is utilized as patches' color similarity to pseudo-background patches. Finally, we propose an optimization framework to combine the two saliency measures. The experiments demonstrate that our method can efficiently highlight salient objects and reduce background noise, which out-performs most state-of-the-art approaches.