Saliency Detection Optimization via Modified Secondary Manifold Ranking and Blurring Depression.

We propose an unsupervised saliency optimization method mainly via modified secondary manifold ranking and blurring depression (SMBD). Generally, saliency object is detected insufficiently by most methods. To solve this problem, a modified manifold ranking is circulated twice to detect saliency object completely. A blurry degree detection approach is introduced to locate blurring regions, which is more likely to be background. As a result, blurring regions are depressed by SMBD to avoid mistaking background as foreground. Our method is performed based on hierarchical luminance for better performance. Extensive experimental results demonstrate that SMBD is able to promote the performances of state-of-the-art saliency detection algorithms significantly.