Abstract: Aiming at the limitation of Expectation Maximization(EM) algorithm for mixture model parameters, this paper presents a fuzzy constrained mixture model for image segmentation. According to the mixture model based on the precondition of pixel independence, it solves the parameters of model by Expectation Maximization(EM) algorithm. The pixel spatial information with the fuzzy method is introduced to correct the independent assumption of the pixel and reduces the influence among the parameters of mixture components. Experimental results show the algorithm does not add model parameters. It needs a model selection criterion to choose suitable number of mixture components.

Key words: image segmentation, mixture model, Expectation Maximization(EM) algorithm, fuzzy constraint, model selection

摘要： 医学图像分割中的期望最大化(EM)算法在求解混合模型参数时存在局限性。为此，提出一种模糊约束的混合模型图像分割算法。该算法以像素的独立性假设为前提，在采用EM算法对模型参数进行求解的过程中，通过模糊集合论方法，引入像素空间信息。实验结果表明，该算法没有引入新的模型参数，能够保持独立混合模型的简单性，且具有自动模型选择能力，可以获得较理想的分割结果。

关键词: 图像分割, 混合模型, EM算法, 模糊约束, 模型选择

CLC Number: 

• TP391