Salient Object Detection Using Window Mask Transferring with Multi-layer Background Contrast

In this paper, we present a novel framework to incorporate bottom-up features and top-down guidance to identify salient objects based on two ideas. The first one automatically encodes object location prior to predict visual saliency without the requirement of center-biased assumption, while the second one estimates image saliency using contrast with respect to background regions. The proposed framework consists of the following three basic steps: In the top-down process, we create a specific location saliency map (SLSM), which can be identified by a set of overlapping windows likely to cover salient objects. The binary segmentation masks of training windows are treated as high-level knowledge to be transferred to the test image windows, which may share visual similarity with training windows. In the bottom-up process, a multilayer segmentation framework is employed, which is able to provide vast robust background candidate regions specified by SLSM. Then the background contrast saliency map (BCSM) is computed based on low-level image stimuli features. SLSM and BCSM are finally integrated to a pixel-accurate saliency map. Extensive experiments show that our approach achieves the state-of-the-art results over MSRA 1000 and SED datasets.