Salient object detection by robust foreground and background seed selection

Abstract

This paper presents a novel background and foreground seed selection method for graph-based salient object detection. First, according to the boundary prior which considers that the image boundary is mainly the background, we select the initial background seed set and optimize it through our proposed two-stage background seed correction processes by combing multiple internal image features. Second, different from most existing center-based prior methods, this paper uses the convex hull of the point of interest to estimate the location of salient objects and thus obtains the initial foreground seed set. Moreover, foreground seeds are refined by the proposed foreground seed correction process, which depends on the color and spatial differences between seeds. Third, we adopt the extended random walk to propagate the background and foreground labels. Finally, a fusion model is proposed to integrate the background- and foreground-based saliency maps, generating final salient object detection results. Experiments on publicly available data sets show that the proposed algorithm achieves better results in contrast to other state-of-the-art methods.