Robust Ranking on Manifold for Salient Object Detection

Saliency detection is to find the most important object automatically according to the human visual in the unknown scene. Most existing algorithms detect the salient object using various salient object features. In this paper, we present a novel saliency detection method by an iterated graph Laplacian based ranking on manifolds to determine whether the region is salient or not. Firstly, we segment the input image into several regions, and then compute the ranking function based on a robust graph Laplacian. Secondly, we estimate each region's saliency value using the background and foreground queries respectively. The background queries have determined using analysis of the image edge feature. The foreground queries have produced using the concept called boundary connectivity. Experimental results prove that the proposed algorithm outperforms many of the recent state-of-art and classical algo-rithms.