Manifold Ranking-Based Kernel Propagation For Saliency Estimation

Saliency estimation becomes a hot research topic due to its wide and successful application in almost all vision related problems. However, it is still far from satisfactory in saliency estimation techniques due to the complex visual content and various requirements. In this paper, we propose a manifold ranking based kernel propagation (MRKP) approach for visual saliency estimation. MRKP begins to work on background seeds for manifold ranking on four image boundaries individually and select representative salient seeds. Pairwise constraints of must-link and cannot-link are formed with the boundary background seeds and selected salient seeds. Then, pairwise constraints guided saliency seed kernel learning and saliency kernel propagation are sequentially conducted in MRKP to estimate visual saliency of images. Experimental results demonstrate that the proposed MRKP has a good ability of learning discriminative kernel structure for saliency estimation.