Research On Image Segmentation Based On Level Set Methods

Image segmentation is a basic problem in the current digital Image processing and Pattern Recognition. In the past decades, the research about this always has been the scientific issues of concern. Image segmentation purpose is that find the region of image from the destination image. Therefore, it's better to analyze the problems. Or to analyze the local image more comprehensive. As the basis of image segmentation and application breadth, there are many image segmentation algorithms. The algorithm base Partial Differential Equation and Level set methods is a hottest research in the current society. Partial Differential Equation is the important problem in scientific computing. Although it which introduction to Partial Differential Equation increase the mathematical complexity of digital images increases, it makes the principles and effectiveness of segmentation mutations, as the alchemy into modern chemistry. Level set method is the important resolve in Partial Differential Equation. It could auto handled the topology structure. And changes to the interface energy by means of implicit functions into a Hamilton-Jacobi partial differential equations and numerical solution method used to solve it.The thesis based on the analysis and synthesis of some of the existing home and abroad some algorithm, the level set method and PDE based image segmentation techniques are further discussed and the C-V level set model and the level set initialization problem proposed improvements and gives the comparison of theoretical and experimental results, where the comparison also includes the widely used switched compare the results obtained. The main contents of this thesis are as followers:1.We gives a semi-automatic intelligent scissors algorithm, this method is because the interaction with the people (though few) and therefore complex image segmentation applications have high robustness, it has been involved in my project reference, in this article for the purpose described on the following algorithm results in a more automatic2.Chan and Vese combine of level-set method and a variation (Mumford-Shah) model, propose the C-V level set model, according to the traditional based on parameters active contour model and geometric active contour model. It is different that the model to extract object boundaries do not rely image gradient information, therefore, for meaningless or edge gradient image blur can be well segmentation. But the model exists and the general level set model calculated as large weaknesses. Paper analysis the massive computation; propose an improved method, which greatly improves the computational time required. The effect of segmentation must also be improved. At the same time, for the presence of multi-channel image in which a single channel can not completely describe the characteristics of the image, we give an algorithm to resolve it.3. Paper introduces a without to re-initialize the level set evolution method, according to the classic methods, this methods on calculating the level set function of the energy functional to add an internal energy term, which is used to correct the level set function the metric (distance function symbols) bias exists. The method can not overcome the traditional geometric active contour model; the flaw is in the classical level set method on a major breakthrough. The paper analysis on initial position on sensitive issues, and found that the existence of this method can not change the direction of adaptive problems; gives a weight coefficient by the variable to solve this problem, improving the technology while on the curve stop function, slow convergence, gives a fast convergence function. Paper gives way to not only improve the speed and get a better result. For a class of uniform intensity image segmentation results of a comparative analysis of the local constraint by adding items, using the existing methods, thus proving the validity of our approach.