Research On MAP-based Segmentation Method For Images With Intensity Inhomogeneity

Segmentation of medical image with intensity inhomogeneity plays a fairly important role in modern medicine for clinical diagnosis and treatment. Geometric Active Contour(GAC) based on level set method has been proposed to solve this problem and got some achievements. The popular geometric active contour models generally adopt the idea of local fitting. Some researchers introduced probability theory into the GAC and proposed a successful model called Most a Posterior(MAP) model. In this paper, two novel models based on different ideas are proposed to cope with this hard problem. The major contributions of this paper can be summarized as follows:A novel level set model based on operation of nonlinear median fitting and combined with image edge information is proposed. Compared with traditional model based on operation of linear fitting, this novel model can effectively remove the high-level noises whereas keeping the original edge information of image from over smoothing during the process of image fitting. Experiments on synthetic and real images demonstrate that this model has promising performance in terms of noise robustness, segmentation of images with intensity inhomogeneity and computational cost.A novel level set model based on information entropy measure of intensity inhomogeneity and combined with prior information of threshold segmentation is proposed. After analyzing the sensitivity to strong noise of standard deviation measure of traditional MAP model, a new measure-information entropy is proposed to solve this problem. By constructing the new intensity inhomogeneity measure and combining with the prior information of threshold segmentation, segmentation capability of this model is improved quite a lot. With experiment verification, the new model not only could get similar good results on images with intensity inhomogeneity、high-level noise and texture with the original model, but also solved the problem of sensitivity to strong noise of the original model.