Multiphase Image Segmentation Based On Improved Vese-Chan Variational Level Set Model

Medical Image Segmentation is a traditional and challenging project, and a lot of new segmentation algorithms have been proposed by experts.Level set method describes the evolution of geometric active contour by a compact mode and provides a stable numerical algorithm. The energy function of Mumford-Shah model synthetically uses the information of image's boundary and region, and its contour evolution is not concerned to the gradient of boundary. The model gets a well result in blurred and discontinuous boundary. Chan and Vese introduce the active contour model based on simplified Mumford-Shah model (called C-V method), which can be used both in contour detection without edges. C-V model can well detect the vacuum of object, but have innate drawback when dealing with complicated multiphase images.C-V method is extended to the so-called Vese-Chan variational level set model which needs multiple level set functions to segment multiphase image by Vese and Chan. This method takes good result in dealing with the simple images whose pixels accord mean distribution, while bad result in dealing with the images whose pixels accord complicated probability distribution. In this paper, the extended studies of this model here include general region model beyond its piecewise constant presumption and a general formulation for the multiphase expression. The first extension is based on maximum a posteriori of region partition and some parameters are estimated for images with Gauss and Rayleigh distributions. The second extension is based on the transformation between a natural number and binary numbers. Experimental results show that the extended Vese-Chan model correctly segments images.