Saliency Detection Based on Foreground and Background Propagation

In recent years, image saliency detection has become a research hotspot in the field of computer vision. Although significant progress has been witnessed in visual saliency detection, several existing saliency detection methods still cannot highlight the complete salient object when under complex background. For the purpose of improving the robustness of saliency detection, we propose a novel salient detection method via foreground and background propagation. In order to take both foreground and background information into consideration, we obtain a background-prior map by computing the dissimilarity between superpixels and background labels. A foreground-prior map is obtained by calculating the difference of superpixels between the inner and outer of a convex hull. Then we use label propagation algorithm to propagate saliency information based on foreground and background prior maps. Finally, the two saliency maps are integrated to generate an accurate saliency map. The experimental results on two public available data sets MSRA and ECSSD demonstrate that the proposed method performs well against the state-of-the-art methods.