一种分层自适应快速K-means算法  

Hierarchical adaptive fast K-means algorithm

作　　者：张晓琳[1] 崔宁宁[1] 杨涛[1] 李洁[1] 

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010

出　　处：《计算机应用研究》2016年第2期421-423,427,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61164018)

摘　　要：提出一种分层自适应快速K-means(hierarchical adaptive fast K-means,HAFKM)算法对图像数据库分类聚簇。HAFKM根据提出的分层策略构建一棵非平衡聚类树,通过自适应的方法 CEC(cluster evaluation criterion)确定了除根节点外的每棵子树的分支数目,而在聚类树的每一层聚类中使用一种提出的判别函数(cost-function)在颜色直方图上根据颜色等级直接聚类,从而可以在整棵树上快速聚类。实验表明,HAFKM算法通过在非平衡树上逐层聚类,并且通过CEC准确判断聚类数目,可以快速、高效地实现数据库的分类聚簇。This paper put forward a method of hierarchical adaptive fast K-means （HAFKM） classifying and clustering for im- age database. According to the proposed HAFKM algorithm, this paper built a unbalanced clustering tree, and determined ev- ery subtree branch number except the root node, through the method of adaptive CEC （cluster evaluation criterion）. In each layer of the clustering tree it used a proposed cost-function to cluster on the color level histogram directly, and then could fast cluster database in the whole tree. Experiments show that HAFKM can cluster layer by layer in the unbalanced tree, and through CEC determine the correc tnumber of clusters. It can realize the classification of database fast and efficiently in the end.

关 键 词：HAFKM K—means算法 分层聚类 自适应 大数据库 聚类树 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]