Saliency Detection Based On Hierarchical Information Fusion And Random Walk

In recent years, fueled by the popularity of low-cost digital image capturing devices, the scale of digital images and videos have almost the exponential growth.How to find the required image from a big database efficiently has become a hot research problem in computer vision. The saliency detection method which is based on simulating human’s visual attention, is an important way to help computer sensors to understand the world.In this paper,we propose two novel saliency detection methods. Most of existed saliency algorithms can only detect the interesting pixels or regions. In order to distinguish the saliency object from the complex background region and guarantee the uniformity of patches in the same object, a new method based on hierarchical information fusion is proposed in the third chapter. It is different from the state-of-art method that the mid-level superpixels and object-level regions are used to adjust the raw saliency map in our method. Firstly an edge-preserving filtering is adopted as a pretreatment, and then the mid-level superpixels are generated by SLIC algorithm.Secondly, the mid-level raw saliency map is obtained by saliency filter, and will be adjusted by two priors, which can reduce the impact of complex background regions.After that, the mid-level superpixels are clustered to object-level segments by spectral clustering, and an object boundary prior is defined to enhance the consistency of the saliency map. Finally, the saliency label will be diffused from superpixels to object-level regions by heat diffusion.Based on the good description ability of random walk process for human visual attention, a novel visual saliency detection method based on the random walk is proposed in the fourth chapter. Compared with the traditional method, this method has the following two aspects of contributions. First, compared with the ordinary random walk, the proposed method in this paper can effectively convergence to the steady state. Secondly, the method in this paper is more reasonable and more robust which adopt round trip time of visual transfer for saliency detection. In this chapter, we give a large factor to the background and foreground seeds. Then we evolution the lazy random walk process on the undirected graph generated by image super pixels. Finally,we obtain a robust visual saliency detection result.Both qualitative and quantitative evaluations on the MSRA-1000 database demonstrate the robustness and efficiency of the proposed method against several state-of-the-art methods.