利用组合马尔可夫模型的非监督图像分割方法

提出了一种基于边缘辅助的组合马尔可夫随机场模型(E-CMRF),并应用于非监督图像分割.在传统的马尔可夫标号场(MRF)基础上引入边缘二值随机场,二者相互作用构成组合随机场.该模型使用期望最大(EM)算法对待分割图像完成参数估计,并运用动态能量权值提高收敛速度.最后根据贝叶斯定理将图像分割问题转化为最大后验概率的求取,运用改进的Metropolis采样算法求得最大后验概率解.实验结果证明,该分割方法不需要人工给出先验信息,在具备抗噪性等特点的同时提高了分割精度.

Unsupervised graph cuts via compound Markov random field model

A novel edge-based compound Markov random field(E-CMRF) model was proposed,which has been used in unsupervised graph segmentation based on image features.E-CMRF combines label field with edge field to construct a compound Markov random field.Expectation maximization(EM) algorithm was adopted to estimate the parameters and dynamic weighting parameter between the two energy components is introduced for fast convergence.A modified Metropolis sampler algorithm was developed to maximize the posteriori conditional probability distribution based on the Bayesian theory.Experiments on synthetic images and natural images showed that the proposed model is able to produce more accurate segmentation results and also less sensitive to noise than some of the existing models in common use.