%0 Journal Article
%T Cluster analysis methods based on Fourier transform and graph theory<br>基于傅里叶变换和连通图的聚类分析方法
%A JU Yu-fang
%A LEI Xiao-feng
%A DAI Bin
%A ZHUANG Wei
%A SONG Feng-tai
%A <br>巨瑜芳
%A 雷小锋
%A 戴　斌
%A 庄　伟
%A 宋丰泰
%J 计算机应用研究
%D 2012
%I 
%X Clustering is to find the best partition of unlabeled observations under a certain group structure hypothesis. For the shortcomings in the existing clustering algorithms, this paper assumed that the results of the data sample was intensive and the differences among every cluster were significant. Based on the assumption it presented a cluster analysis method called FGClus based on discrete Fourier transform and graph theory. First, this method calculatd k-distance matrix of each sample point as a sequence of the input signal of discrete Fourier transform, then extracted the minimum amplitude of the complex frequency domain items and constructed the input sequence of inverse Fourier transform, to get the optimal threshold value of the space in the time domain. Finally, it used threshold and connected graph to guide the final clustering process. Large numbers of experiments show that FGClus algorithm can overcome existed shortcomings of K-means algorithm, such as the number of clusters must be determined before clustering, the results is sensitive on initial selection of representative points and it just can cluster spherical datas, which achieves good clustering results.
%K cluster analysis
%K discrete Fourier transform
%K connected graph
%K shortest path KNN query
%K optimal threshold<br>聚类分析
%K 离散傅里叶变换
%K 连通图
%K 最短路径K近邻查询
%K 最佳阈值
%U http://www.alljournals.cn/get_abstract_url.aspx?pcid=5B3AB970F71A803DEACDC0559115BFCF0A068CD97DD29835&cid=8240383F08CE46C8B05036380D75B607&jid=A9D9BE08CDC44144BE8B5685705D3AED&aid=C2AB4F886AE798C970BFF818873AD1EF&yid=99E9153A83D4CB11&vid=771469D9D58C34FF&iid=5D311CA918CA9A03&sid=A66F9C7240EBF5BE&eid=2CE26DEC7508C852&journal_id=1001-3695&journal_name=计算机应用研究&referenced_num=0&reference_num=12