Abstract: Compared with Fuzzy C-Means（FCM） clustering, Possibilistic C-Means（PCM） has a better anti jamming capability. But the Possibilistic C-Means clustering is very sensitive to initial conditions and is very easy to cause the clustering result of consistency. And it doesn’t take into account the pixel spatial information. It is extremely unstable when it is used in image segmentation especially in multi-object image segmentation. Based on the PCM clustering, the prior spatial constraint is incorporated according to Markov random field theory, to build a new clustering objective function including the establishment of gray information and spatial information. This paper presents a new image segmentation algorithm（MPCM） combining Markov and PCM clustering. With experiments, using MPCM algorithm can achieve a better segmentation result than PCM in multi-object image segmentation.

Key words: image segmentation, Possibilistic C-Means（PCM）, Markov random field, clustering

摘要： 与模糊C均值（FCM）算法相比较，可能性C均值（PCM）聚类算法具有更好的抗干扰能力。但PCM聚类算法对初始化条件很敏感，在聚类的过程中很容易导致聚类结果一致性，并且没有考虑到像素的空间信息，用在图像分割尤其是多目标图像分割上效果极不稳定。在PCM算法的基础上，利用Markov随机场中的邻域关系属性，引入先验空间约束信息，建立包含灰度信息与空间信息的新聚类目标函数，提出马尔可夫随机场与PCM聚类算法相融合的图像分割新算法（MPCM算法）。实验结果表明，在多目标图像分割上利用MPCM算法可以取得比PCM更好的分割效果。

关键词: 图像分割, 可能性C均值, Markov随机场, 聚类