基于核函数距离测度的加权模糊C均值聚类与Markov空域约束的快速鲁棒图像分割

由于模糊C均值聚类算法（FCM）存在两大缺陷：（1）对于非球形的数据聚类形式鲁棒性不够；（2）只考虑图像中的数值特征信息，而忽略了像素间的空间约束关系，因此FCM算法在对含有噪声的图像进行分割时缺乏足够的鲁棒性。针对以上问题，本文提出了一种结合Markov空域约束与基于核函数距离测度的加权模糊c均值聚类的快速鲁棒图像分割方法。为克服缺点（1），我们使用基于核函数的距离测度取代FCM中的欧氏距离，并使用加权模糊聚类的方式保证了计算的简洁性。与此同时，我们用Markov随机场描述图像的空域约束信息，并且通过数据融合的方法将模糊分割结果与空域约束信息结合在一起，从而得到既包含像素数值特征又包括空域约束信息的图像分割场。这样既克服了传统模糊C均值聚类算法的缺点，又最大限度地保证了分割算法计算的简单有效性。

Fast Robust Image Segmentation Based on Weighed Fuzzy C-Means Clustering with Kernel-Induced Distance Measurement and Markov Spacial Constraint

In image segmentation,Fuzzy C-Means Clustering (FCM)laeks enough robustness to noise owing to:(1)the non-robustness of Euclidean distance for non-spherical structure of input data:(2)disregard of spacial constraint infor- mation.To overcome the problems of FCM,a fast robustness image segmentation algorithm based on the Markov spa- cial constraint and weighted FCM with kernel-induced distance is proposed in this paper.To overcome the first problem of FCM and keep the computation simplicity,we replace Euclidean norm with kernel-induced distance and get the fuzzy partition result with weighted fuzzy clustering.Then,the Markov random field is used to represent the space constraint information of an image.Based on the data fusion of Markov constraint information and the fuzzy segmentation result, the new algorithm overcomes the problems of FCM and keeps the computation simplicity.