基于局部空间自适应MRF模型的图像分割

针对固定参数的点对马尔可夫随机场(Pairwise MRF)模型不能充分描述图像丰富的统计特征的问题,在研究 Pairwise MRF 模型的基础上,提出一种自适应分割算法.该算法首先建立一种空间自适应的局部区域 MRF 分割模型,并对局部区域的先验知识进行自适应估计;然后通过局部收敛的循环置信度传播(LBP)算法最大化自适应 MRF 模型的全局后验概率.实验结果表明所提出算法具有较好的分割结果.

Image segmentation based on local spatial adaptive Markov random field model

An adaptive segmentation algorithm is proposed based on local spatial adaptive Markov random field(MRF)
model to solve the problem that the Pairwise MRF model is insufficient to capture the rich statistics features of images by a
single set of fixed parameters. Firstly, the proposed algorithm constructs the local spatial adaptive model based on Pairwise
MRF. For every local region, the local prior is adaptively estimated. Then loopy belief propagation(LBP) algorithm based
on the local region belief convergence is used to maximize the global posterior probability of the adaptive MRF model.
Experimental results show that the proposed algorithm can provide a better segmentation result.