基于可能性FMRF的红外图像分割算法及其参数估计

Image segmentation and parameters estimation based on fuzzy Markov random field with possibility theory

模糊马尔可夫随机场在多值图像分割问题中多采用分段模糊的方法将多值问题转化成多个两值模糊问题,这种方法的模糊度完全依赖像素的灰度值,从而很容易陷于局部最优解。基于可能性测度的模糊随机场摆脱了隶属度对灰度的依赖,使分割结果更容易收敛于全局最优。同时基团类型"相似性"的提出,改进了随机场基团定义的苛刻性,使得基团类型具有更好的包容性与多样性,可广泛地应用到复杂环境下的多值图像分割问题中。最后给出了该算法的EM参数估计方法和图像分割仿真实验。

Fuzzy Markov random field （FMRF） is an efficient model for data clustering or image segmentation. An improved FMRF segmentation method based on the possibility theory is presented. The originality of this algorithm based on the fact that pairwise fuzzy model is incapacious to describe the correlations between different classes in multi-level segmentation.Possibility theory is just a shortcut for this kind of multi-level uncertain classification, which could give more flexible searching ability. The possibility theory is introduced into FMRF to strengthen abilities for searching optimum and the potential energy definitions for similar cliques is proposed. Firstly, definitions about possibility theory in improved FMRF are given. Then, EM algorithms are used to estimate unknown parameters. Finally the experiments demonstrate that the algorithm is efficient to distinguish fuzzy edges or mixed areas.