Hierarchical saliency: A new salient target detection framework

Simulating the shift character of visual attention, we propose a novel concept of hierarchical saliency and develop a detection framework. First, a given image is over-segmented into coarse and fine layers which respond to two scale superpixels. Then, we estimate the saliency maps from coarse to fine. In the coarse layer, we present a new self-adaptive algorithm to construct the superpixels graph, employing the manifold ranking approach to optimize it. In the fine layer, sparse reconstruction is used to obtain the saliency regions. At last, we propose a Restricted Voting Strategy (RVS) to fuse two layer saliency maps into one hierarchical saliency map. Different from the prior methods, the targets of the final map are labeled layer-wise. The final result can be directly applied to more high-level computer vision tasks in various situations. For the requirement of hierarchical saliency evaluation, we construct the CAS-HAS dataset. We exhaustively evaluate the framework on the proposed data set and three benchmark data sets. The experiment performance is comparable with the sate-of-the-art approaches.