The Improvement On The Image Segmentation Models Of CV And GAC Based On Partial Differential Equation

Images are essential information carriers in real world and the processing and relat ive studies on them are of great importance. The image segmentation as the key part of image processing should never be ignored. Though there are many traditiona l methods concerning image segmentat ion, they have certain limitat ions and lack a strong theoretical support.Partial different ial equation, when used for the image processing, not only enjoys powerful theoretical support but is in coordination w ith those classic image theories. The introduction of level set method has attracted more interest in this field and the evolution of le vel set function can totally make up the defic iencies of the opaque curve evolution based on differential geometry.In this paper, we have not only conducted further study on the partial differential equation dealing w ith the image segmentation, but summarized the development of level set method and the changing models. We lay emphasis on two classical segmentation models: GAC model and C V model, both of which stand out as the landmarks of boundary- based and region-based. Their significant advantages, such as anti- noise abilit y, in the filed of image- processing enable them to adapt the changes of topology on the basis of leve l set method. Besides, they set off strengt hs of theory and practice by giving a clear explanation on the relat ionship between them. However, as a result of the complexit y of images and the defects of models, inevitably, the two models have some shortcomings.GAC model is sensit ive to initia l contour curve and cannot segment boundary of the depression. C V model behaves less satisfying on the effective segmentation in the images wit h non- homogeneous regions. Based on those defects mentioned above, this paper proposes improved models.Guided by LIF model, we have improved the CV model. Except extracting local features by introducing w indow funct ion, we also illustrate the differences between steep region and flat region with the help of variation changes. By doing this, simple arithmetic mean in C V model can be replaced through successive weighing. The latest model is based on local informat ion instead of on global informat ion, which can guarantee the image segmentation to comply w ith the general image segmentation rules.In this paper, a variable coeffic ient term is proposed for the GAC model, which can be used to solve the segmentation of the concave boundary, which is different from the non boundary region and the boundary region. At the same time, to a large extent, the limitation of the original model boundary leakage is alleviated. Since the origina l model is highly dependent on the init ial contour curve, the terms of the motion direction of level set function are extracted from region- based models and then transplanted to GAC model in order to solve the problem.By applying finite difference method, this paper has constructed numerical format on the improved model above. In addit ion, relat ive segmentation experiment has proved the effic iency of this model which owns those advantages that the original model lacks.