Edge-based method for detecting salient objects

Abstract

A novel edge based method for detecting a salient object in an image is proposed. The core idea of our work is that the problem of detecting a salient object is transformed to the problem of finding curves in the edge map. It requires only edge detection and closed curve detection. The novelty lies in two aspects: first, when producing an edge map, the number of edge pixels is limited to make sure that edges detected in the background are relatively little; second, when detecting a salient object from an edge map, the contour of the salient object is detected in a simple way to decrease the effect of edges in the background. Experiments conducted on Liu et al.'s [Proc. CVPR, pp. 1-8, Minneapolis, MN (2007)] dataset of 5000 images show that the method is comparable with several popular methods from literature.