Research On Level Set Image Segmentation Method Based On Chan-Vese Model

Chan-Vese model is a geometric active contour model based on regional information. It does not depend on gradient information, so it can segment image with noise and weak edge effectively, and combining the level set method for the numerical solution, it can naturally change the topology structure in the contour evolution process, so it has aroused widespread attention. But Chan-Vese model has its defects,this dissertation conducted the research and improvement of the deficiencies of Chan-Vese model. This dissertation is as follows:(1) The Chan-Vese model is a non-convex functional. Seeking an extreme value for the functional can only obtain a local optimal solution. So the application of Chan-Vese model for image segmentation is difficult to obtain an ideal result in the global range. For improving this problem, a new method for segmenting images is proposed in this dissertation, which is based on Chan-Vese model combined with GVF.GVF spreads the edge gradient information to the entire image and it guides the evolution of Chan-Vese model to the correct target edge in the global range. The improvement retains the basic advantages of Chan-Vese model. The experimental results show that the segmentation result and convergence speed of the proposed method in this dissertation are obviously better than the traditional Chan-Vese model,and the improvement is feasible.(2) The Chan-Vese model divide region to be segmented into the internal region and the external region in the evolution process. So the application of Chan-Vese moldel for multi-gray image segmentation can only obtain the two phase segmentation result. To solve the problem, a multi-gray image segmentation method is proposed that based on single level set serial evolution in this dissertation. Different from the traditional single level set tree-like model segmentation method, This method only fit gray average value to the internal region of the active contour, the one way serial segmentaion of multi-gray image is performed by single level set multiple evolution. A level set function can be used to generate the initial active contour in the specified region. And the Mean-Shift algorithm is used to judge the target of region to be segmented and achieve the unsupervised automatic segmentation. The experimental results show that the segmentation effect of the proposed method in this dissertation can effectively reduce the number of judgement and total time of segmentation in the segmentation process, and therefore the method proposed in this dissertation is feasible.