叶海军

摘 要:从传统目标函数聚类方法的思想出发,在基于样本集统计特征的基础上,提出基于统计特征加权模糊C-均值聚类方法,并提出基于统计特征的权值计算方法。分别利用图像的一维灰度特征与一维灰度统计特征加权和二维灰度特征与二维灰度统计特征加权,将两种特征加权的模糊聚类方法应用于灰度图像二值化,并将该方法的处理结果与其他二值化方法处理结果进行详细的比较。实验结果表明,该方法能够有效地实现图像的二值化。

关键词:统计特征;模糊C-均值聚类;图像二值化;权值

中图分类号:TP391.4文献标识码:B

文章编号:1004-373X(2009)12-099-04

Fuzzy Clustering Method and Its Application Based on Statistical Characteristics Weighting

YE Haijun

(China Academy of Electronic and Information Technology,Beijing,100041,China)

Abstract:A weighting fuzzy C-means clustering method based on statistical characteristics and the weighting value′s calculation method based on statistical characteristics are proposed,which sets out from the traditional target function clustering method.The two kinds of weighting fuzzy clustering method are applied to image binarizatation by utilizing one-dimensional gray characteristic of image and one-dimensional gray statistical characteristic which is weighting value,two-dimensional gray characteristic of image and two-dimensional gray statistical characteristics which are weighting value.The paper shows result comparison of image binarizatation with other image binarizatation methods detailedly.The application examples show that the method can realize image binarizatation availably.

Keywords:statistical characteristics;fuzzy C-means clustering;image binarizatation;weight

0 引 言

模糊聚类分析是多元统计分析的一种,也是无监督模式识别的一个重要分支。模糊聚类分析被广泛应用于模式识别、图像处理、知识发现、计算机视觉和模糊控制等许多领域。模糊聚类分析已有很多方法,在基于目标函数的聚类方法中最具有代表性的是模糊C-均值聚类方法(Fuzzy C-means,FCM)[1,2],最初又称ISODATA聚类方法,它是由Dunn[3]从硬C-均值聚类方法(Hard C-means,HCM) [4]引出的,后又经过Bezdek归纳并加以完善。FCM方法是通过对目标函数的迭代优化实现对给定有限样本集的划分[4]。在当前基于目标函数的模糊聚类分析方法研究中,都是基于这一基本思想而提出的算法。

这里从分析给定样本集合中样本点本身、样本点与聚类中心距离、样本点隶属度和样本点统计特征出发,提出了基于统计特征加权的模糊C-均值聚类方法(Weighting Fuzzy C-means,WFCM),并将该方法应用于灰度图像的二值化。……