Closed contour extraction by perceptual organization and global saliency

In this paper a contour extraction algorithm combing perceptual organization and global saliency is proposed to reduce redundant edges and improve the precision of the extracted contour. The novelty lies in two aspects: firstly, a salient edge detection algorithm incorporating the saliency information is proposed to reduce redundant edges in the background; Secondly, a saliency based cost function is introduced to make the algorithm converge at regions with high saliency, which could better mark the foreground object. Experiments show that the proposed algorithm could improve the accuracy of contour extraction, meanwhile, reduce the computation time efficiently.