Salient Object Detection Algorithm Based on Globally Isolation and Locally Compaction
|Course||Communication and Information System|
|Keywords||Visual attention Saliency Random walk model Markov chain model Graph representation Space compaction Region uniform Entropy of orientation|
Visual attention is can rapidly and accurately allocate the limited resources to visual salient areas. Meanwhile, visual saliency detection has a wide range of application in image compression, target detection and image retrieval. Therefore, how to build machine vision model inspired by the biological are of great concern. The current presented visual attention region detection algorithm is mainly focused on the inter-pixel contrast and lack of global perspective to analyze and understand the salient object, so that the detection result is unreasonable when the input image is more complex. In view of this, some new visual attention computational model are analyzed in this paper.Firstly, a salient object extraction algorithm based on space distribution and local complexity is proposed. First of all, the bright saliency map is obtained by computing the contrast of local area and its multiple scale neighborhoods, and then the color saliency map is computed using conspicuous, space distribution and locally uniform of color information. Meanwhile, the orientation saliency map is obtained by multi-scale analysis of the spatial distribution and regional complexity of the orientation. Finally, an integrated saliency map of the input image is composed through the fusion of bright saliency map, color saliency map and orientation map, and the test results show that the algorithm is feasible.Secondly, an unsupervised graph presentation random walk salient object extraction algorithm based on global isolation and local homogeneity is proposed by formulating salient region detection as Markov random walk. First of all, the graph model is formed by dividing the input image into pixel block and using color and orientation features to determine the weight of edge, and then the global properties and local properties are extracted through the random walk on a complete graph and a k-regular graph. Finally, the saliency map is obtained by combining the global properties and local properties of the object, and the salient object is located and extracted according to the saliency map. The better experimental results are obtained.Finally, an unsupervised salient object detection algorithm with a hybrid graph model is proposed according to biological visual attention mechanism. First of all, a hybrid graph model is formed by exploiting color information and orientation prior. The vertices are connected by undirected edges and directed ones, and the undirected edges represent the color difference between two block images, while the directed edges represent the dependence of the spatial distribution and regional complexity of the orientation feature. Secondly, the saliency of pixel block is measured by random walk model. Finally, the saliency map is obtained by combining the global properties and local properties of the image. Experimental results show that the proposed method is effective.