Visual Saliency Based on Local and Global Features in the Spatial Domain

The human visual system can quickly and efficiently capture the salient objects in a scene. Based on the biological mechanism, a new multi-scale saliency analysis method is proposed in this paper, in which the differences of region colors and spaces are calculated in different scale and their saliency map are fused together. First, we calculate the image saliency by using the color and space information of both local and global in single scale. Then by applying the multi-scale fusion, we can effectively inhibit outstanding but not salient region in each single scale, and different scale can also reflect salient region of the images from different aspects. The experiment results show that this algorithm can effectively predict the salient region attracting human attention. Our method has the state-of-the-art performance and achieves excellent results for salient objects of different sizes and salient region with complicated background in an image.