Saliency detection based on foreground appearance and background-prior

Abstract

Abstract Saliency detection has drawn increasing attention in the communities of computer vision and image processing. In this paper, we propose an effective saliency optimization scheme by taking account of the foreground appearance and background prior. First, a foreground appearance-based saliency is obtained by integrating the uniqueness and compactness of the GMM-decomposed soft abstraction. Second, a foreground-based background seed selection algorithm is proposed to obtain the enhanced background prior based saliency. It can well alleviate the influence of the on-boundary objects to the final saliency in conventional background prior based saliency algorithms. Third, the problem of foreground appearance and background prior-based saliency integration is formulated as a convex optimization problem. The optimal saliency map could be obtained through solving the optimization problem. Extensive experiments on five public databases demonstrate that the proposed method consistently outperforms the state-of-the-art saliency detection methods.