Research On Image Saliency Based On Superpixel

Human beings are quite adept at detecting objects of interest in complex visual scenes with the assist of Visual System(VS). It has been widely studied by researchers to simulate visual systems of human beings in order to detect visually salient regions of images. Researchers build the visual attention mechanism in two ways:Top-Down (TD) and Bottom-Up (BU). In this paper, we achieve saliency detecting based on superpiexl in the BU way.Firstly, we select SLIC superpixel segmentation in this article after analysis and summary of methods of superpixel segmentation. The basic knowledge on image saliency is also given in the paper, including mechanisms of human visual attention, visual characteristics and methods of comparison.Afterwards, several classical models of image saliency detection are analyzed and their advantages and disadvantages are summarized. Then we propose the image saliency detection model based on superpixel according to the thought of Edge-Center deal. In this model, we obtain the superpixel sets after segmenting the original images using the method of SLIC superpixel segmentation. Then we integrate the characteristics of colors and space to analyze the global information. We set some seed points on the boundary to build the saliency map and then estimate and label the salient region after the analysis of four edge saliency maps. Moreover, we can get the final saliency map by integrating the boundary saliency maps with the central regions of the saliency maps.Finally, an experiment is made to test the accuracy and effectiveness of the proposed model and the results demenstrate that the model is effective.