Abstract: As one kind of image segmentation methods, fuzzy C-means（FCM）clustering is an effective and simplest method. As a variant of FCM, Possibility C-means algorithm（PCM） algorithm has better clustering performance and?the?enhanced?interpretability?based?on?the?probability?theory. However, both FCM and PCM are sensitive to the noise and outliers due to the membership degree of constraint. To solve the above problems, a new clustering algorithm called maximum the between-cluster（MPCM） is proposed. Which takes the dissimilar penalty term between centers into consideration, widened the distance between the center of the classes, to achieve the best classification of pixels in the image through the dissimilarity parameters [λ]. The experiment results of synthetic texture image, real image and real images with salt and pepper noise shows that the new algorithm is better than the conventional clustering analysis method on the image segmentation technique.

Key words: image segmentation, clustering, Possibility C-means（PCM）, maximum the between-cluster, texture image

摘要： 在图像分割的多种方法中，模糊C均值（FCM）聚类是最简单有效的。可能性C-均值算法（PCM）作为FCM的同类算法具有更佳的聚类性能和概率解释性，但无论是FCM还是PCM均受隶属度的约束影响使其对噪声点和野值点较为敏感。针对以上问题，提出了一种新的称之为类间极大化的PCM算法（MPCM）聚类算法。该算法考虑了对类间的惩罚，通过调控参数[λ]，拉大类中心之间的距离，实现图像中像素点的最佳分类。给出了人工合成纹理图像、真实图像以及带有椒盐噪声的真实图像的实验，结果表明算法在图像分割效果上优于传统的聚类分析算法。

关键词: 图像分割, 聚类, PCM聚类, 类间距极大化, 纹理图像