Saliency Estimation Model Based on Superpixel and Regions Contrast

Human vision system has a remarkable ability to distinguish salient objects from the complex scenes in real time effortlessly and efficiently. But it is still remains the significant challenge to establish the computational model of this basic intelligent behavior in the fields of computer vision. Based on superpixel segmentation and the regions contrast scheme, one saliency estimation model is proposed in the paper. Firstly, the method of the superpixel lattice segmentation is implemented to generate superpixels for the input image, and then according to the superpixel segmentation, the regions contrast strategy is incorporated in order to calculate the saliency maps of the input image, finally the saliency estimation results are obtained with relatively low computational complexity. Compared with the context-aware saliency method, the better advantage of this saliency estimation model proposed in this paper is that the computational complexity is obviously reduced. Furthermore, the full resolution saliency maps can be achieved using the proposed saliency estimation model in this paper. At the same time, the experimental results also clearly demonstrate that the proposed model for saliency estimation is effective.