http://s11/middle/46e3af5bg7a417172f1aa&690
图 Matlab的减法聚类subclust函数对随机生成数据的定位效果

%m文件
% generate k cluseters data and locate by subtractive clustering
% copyright by sunzhihai@HDU 2012.8.6

function data3clusters
data_n=30;
k=3;    %initialization the number of data and the clusters of data

C=rand(k,2);% rand initialization the centers of clusters

data1=10.*repmat(C(1,:),data_n,1)+randn(data_n,2)./10;
data2=10.*repmat(C(2,:),data_n+10,1)+randn(data_n+10,2)./10;
data3=10.*repmat(C(3,:),data_n+20,1)+randn(data_n+20,2)./10;

data=[data1;data2;data3]; %数据生成
plot(data(:,1),data(:,2),'*'); %绘制生成的数据
hold on
[C,S]=subclust(data,[0.25,0.2]); %减法聚类
plot(C(:,1),C(:,2),'ro') %绘制聚类中心

%给聚类中心绘制椭圆
for l=1:k
    hold on
    PlotEllipse(C(l,1),C(l,2),0.25,0.2,0) %椭圆绘制函数
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

几个问题有待进一步研究：

上下限阈值参数的自适应确定？

邻域半径的确定？

数据尺度及方向能否进一步调整？

能否非线性数据的聚类问题？

感兴趣的同学，欢迎一起讨论。eeszh（@）163.com