一种改进的C均值法初始类划分方法

Improved initial classes partition method of C-means algorithm

C均值法是计算机模式分类中一种重要的动态聚类方法,它主要是利用模式间最小距离原则进行分类,其分类效果受到模式的协方差阵和初始类心选取方法的影响。在分析这两个影响因素的基础上,给出了一种新的基于二分或二倍类内距离试探的初始聚类中心的选取和类的初始划分方法。此方法可以不需要先验知识,这对于从事相关研究的人员有一定的参考价值。

C-means algorithm is an important dynamic clustering method in pattern classification. The patterns are classified by the principle of the minimum distances among patterns. The clustering effect of the C-means lies on the covariance matrix of the classes and the selecting method of the initial clustering centers. On the basis of the analysis of the two facts, a novel dimidiate or duple pattern distances seeking method-based selecting way of the initial clustering center and a preliminary partition method are brought forward. It doesn't need preexistent knowledge. This work can help the researchers who are engaged in the relative study.