Image Segmentation Based On Shape Piror Constrained

Image processing technology has been paid considerable attention and in recentyears and it has rapidly penetrated into all aspects of human life and socialdevelopment. Image segmentation is the critical step from image processing to imageanalysis, and then image understanding. Partial differential equation method,especially geometric contour model based on level set, topology changes such asspliting and merging are processed naturally with good results. Level set imagesegmentation based on shape prior is an important research direction of imagesegmentation.In this dissertation, we mainly focus on level set image segmentation based onshape prior. On the base of the analysis of the development process and researchprogress of image segmentation, the theory of curve evolution model, active contourmodels and level set is firstly introduced.Then the classic single prior shape constrained Chan-Vese(CV) segmentationmodel is discussed. But this model loses the ability of segmenting multiple targets atthe same time. So label function is proposed to ensure multiple shape priors can play arole in the evolution of the curve so that shape prior constraint and multi-objectsegmentation simultaneously can both be achieved.At last, shape prior constrained Kernel Principal Component Analysis (KPCA) levelset model with parameter adaption is proposed. Before performing segmentationprocess, KPCA method is used to extract the nonlinear features of training set, soiterative curve can evolve to an arbitrary shape of training set instead of a fixed shape.The model is not only robust to many kinds of interference that appears in the image,but also peform better in the segmentation of objects whose attitude changes greatly.Respect to the fixed shape prior model, this algorithm is more accurate. And theparameters adaption process improves the efficiency when selecting suitableparameters, which improves the segmentation speed a lot.