High-Level Multi-difference Cues for Image Saliency Detection.

Salient detection approaches mainly use single local cues or global cues as its inputs features to detect salient objects, which are sensitive to complex background, so the effect of detection were not satisfactory. In this paper, we investigate the traits of saliency detection and observed the two following facts: Firstly, high-level saliency cues achieve better saliency detection results than low-level saliency cues. Secondly, multi-difference cues achieve better saliency detection results than single difference cues. Based on deeply analysis, we proposed an image saliency detection algorithm through high level multi-difference cues (HMDS). By using multi-difference, not only HMDS could remove the non-salient region effectively, but also it could enhance the pixel value of salient region at the same time. In order to evaluate the performance of HMDS, the proposed method is compared with seven state-of-the-art algorithms on five popular datasets. The final experimental results show that the proposed method performs effectiveness, and will have a perfect application prospect.