Visual saliency detection based on gradient contrast and color complexity

Abstract

Visual saliency detection has been widely researched in recent years. In order to improve the accuracy of the salient object detection, a novel visual saliency detection algorithm based on gradient contrast and color complexity was proposed in this paper. It divided the input images into multi-scale decompositions by Gaussian pyramid firstly. After extracting the gradient contrast and color distribution on the sub-images of different scales, then an efficient fusion method was used to combine them into feature maps. Finally, all the feature maps of different scales were combined to obtain the major visual saliency map by normalization. We conduct comparative evaluations of exiting representative saliency detection algorithms on two benchmark public databases, and the result shows that the proposed algorithm can accurately extract the salient regions, yielding higher precision and better recall rates. In addition, compared with other methods, it does better in highlighting the internal details of salient regions.