Fast Salient Object Detection Based on Segments

In this paper we propose a novel salient object detection algorithm based on segments, named SODS (salient object detection based on segments). We first segment an input image, and then extract a set of features including multi-scale contrast, center-surround histogram, and color spatial distribution based on segments to describe a salient object locally, regionally, and globally. These three features are then combined linearly to get a saliency map to represent the salient object. We validate our approach on two public datasets. Experimental results prove that our method is much faster, more robust and accurate than existing salient object detection methods.