Particle Swarm Optimization Based Salient Object Detection for Low Contrast Images

Abstract

Saliency detection has attracted increasing attentions in computer vision. Although most traditional saliency models can effectively detect the salient objects in natural images, it is still a burning problem in low contrast images, for low lightness and few color information limit the applicability of these models. Different from conventional models, which are not robust on weak light environments, the proposed method uses the particle swarm optimization (PSO) algorithm to estimate the image saliency. First, the covariance feature is used to compute the local saliency of each superpixel region. Then, the PSO search is executed to measure the image saliency in a global perspective. Finally, the graph model is constructed to optimize the saliency value. As the proposed model incorporates both local and global cues, the generated salient objects have well-defined boundaries and uniform inner regions. Experimental results show that the proposed salient object detection model yields better results than eleven state-of-the-art saliency models on low contrast images.