Background aware saliency detection

Abstract

Saliency detection can find important target of scene to improve the subsequent processing speed, so it has important applications in image and video processing fields. But almost all the saliency detection methods depend on the differences between target and background, i.e., the target priors. In this paper we provide a new method using the background priors. Firstly, we decompose an image into compact elements that abstract unnecessary details. Then the initial saliency map is computed with background priors. Finally the multi-scale information is fused to promote the detection performance. Experimental results on a large benchmark database demonstrate the proposed method performs well.