An improved manifold ranking based method for saliency detection
Xiabao Wu, Xiao Lin, Linhua Jiang, Dongfang Zhao · 2017
This paper present an improved Manifold ranking based saliency detection method. Our method constructs a similarity matrix to represent the connection between each superpixel of image. To achieve the optimization, we select some foreground regions as the foreground labels, by using the objective likelihood map technique, and a part of boundary regions as background labels, by using color distinction measure. Based on these prior information, we generate the rough results by manifold ranking algorithm, and merge the results, obtaining our final saliency map. To verify the robustness of our proposed algorithm, we conduct extensive experiments.