一种基于HMRF模型的无监督图像分割算法  

An Unsupervised Image Segmentation Algorithm Based on HMRF Model

作　　者：朱国普[1] 曾庆双[1] 屈彦呈[1] 王常虹[1] 沈博昌[1] 

机构地区：[1]哈尔滨工业大学空间控制与惯性技术研究中心,黑龙江哈尔滨150001

出　　处：《电子学报》2006年第2期374-379,共6页Acta Electronica Sinica

摘　　要：研究了基于隐马尔可夫随机场(HMRF)模型的无监督图像分割问题.对于每一阶模型的图像分割,该算法充分利用了相邻模型之间的相关信息,由此,该算法克服了均值场算法对初始化条件要求非常苛刻的缺点.而且,针对无监督图像分割的模型选择问题提出了带惩罚项的误差平方和阶次判定准则.实验结果证实本文提出的阶次判定准则优于伪似然信息准则(PLIC),并且,该算法具有满意的分割结果.This paper presents a novel unsupervised image segmentation algorithm based on hidden Markov random field （HMRF） model. For each order model segmentation the proposed algorithm makes use of the correlated information between adjacent models. Therefore the algorithm avoids the drawback about that mean field algorithm is restricted by initial condition. Furthermore, in order to solve the model selection problems of unsupervised image segmentation, the sum of squared error criterion with penalty term is proposed. The experiment results testify that the proposed criterion is superior to the Pseudo - likelihood Information Criterion （ PLIC）, and it is shown that the performance of the segmentation is satisfied.

关 键 词：图像分割 HMRF模型 均值场算法 模型选择 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]