基于HMRFEM框架的Byes图像分割方法研究Research on image segmentation method based on Bayes theorem of HMRFEM framework.docx

上海交通大学博士学位论文STUDY 上海交通大学博士学位论文 STUDY oN BAYESL气N IMA(}E SEGMENm气TION BASED ON HA／Ⅱ{F．EM FRAMEWORK ABSTRACT In the MAP—MRF framework,Bayes image segmentation based on MRF model contains two procedures：parameter estimation and pixel classification．However，the current～球下．MAP framework for Bayesian image segmentation has paid more attention to the pixel classification and has not explicitly provided a systematic method for its parameter estimation． Most reported methods have used the MRF as a prior knowledge in the finite mixture model(FMM)to build the M睑P estimation．The prior knowleage has been heuristic based，which ignores the relation between the likelihood model and the prior model．In contrast,the Hidden Markov Random Field model and the expectation．maximization(HMRF—EM、 method provide a formal parameter estimation method within the MRF．Ⅳ队P framework． Howeveg several drawbacks still exist when applying the HMRF—EM framework to image processing．Firstly,the H～ⅡtF-EM framework only works well for those images whose scale changes are not significant．It is caused by its homogeneous assumption．To remedy this drawback,a more flexible framework is required which can deal with the images with significant scale variation．Secondly,the current H^ⅡtF-EM framework iS single scale．Obviously，it iS valuable to extend it to the multiscale framework． In order to achieve a generN HⅣ吸F．EM framework．this dissertation presents following research results： Firstly’recent developments in the Bayesian image segmentation based on Markov model are reviewed．Some related fimdamental knowledge on the MRF model are alSO introduced．In particular,tlle 第1II页 上海交通大学博士学位论文influence 上海交通大学博士学位论文 influence of some factors such as initial condition,the number of iteration, prior parameter and updating mechanism on segmentation performance is discussed in detail． Secondly,the Bayesian image segmentation based on～瓜F and EM algorithm is formulated as a general HMRF—EM framework．办e parameter estimarion，espe