MATLAB并行实现Kmeans聚类算法

使用MATLAB语言实现Kmeans聚类算法，并用并行工具箱进行加速运算。

function [cluster,label] = calculateDistforGpu(data,centeroids,max_iter)
[m,n]=size(data);
fprintf('m=%d,n=%d\n',m,n);
% 获得类别个数
k = length(centeroids);
fprintf('聚类中心个数:%d\n', k);
% 类别标记
%--------------------gpu---------------------%
label = gpuArray(zeros(m,1));
% 得到每个聚类实例
cluster=cell(k,1);
centeroids_copy = centeroids;
for iteration=1:max_iter
    % 将与每个聚类中心最近的实例归为一类
    fprintf('迭代:%d\n',iteration);
    for i = 1:m 
        d = zeros(k,1);
        for j = 1:k
          d(j)=sum((data(i,:)-centeroids_copy(j,:)).^2) ; 
        end
        % 判断每个实例与哪个聚类中心最近，将此实例归为聚类中心所代表的类
        indexes = find(d==min(d));
        % 更新类别标记
        len = length(indexes);
        if len==1
            label(i)=indexes;
            index = indexes;
        else
            index = randperm(len,1);
            label(i)=indexes(index);
        end
        % 为每个类别添加实例
        cluster{index}=vertcat(cluster{index}, data(i,:));
    end
     % 更新聚类中心
     newcenteroids = updateCenteroid(cluster,n);
     % 如果新的聚类中心和上一次的聚类中心相同
     % 则停止迭代
     if sum(sum(newcenteroids==centeroids_copy))
         fprintf('终止迭代次数是:%d\n',iteration);
         break;
     else
         % 替换聚类中心
         centeroids_copy = newcenteroids;
         %-----------------gpu---------------%
         label=gpuArray(zeros(m,1));
         cluster=cell(k,1);
     end
end

% 更新聚类中心
function newcenteroids = updateCenteroid(cluster,dim)
k=length(cluster);
newcenteroids=zeros(k,dim);
parfor i=1:k
    % mean默认按照每一列求均值
    newcenteroids(i,:)=mean(cluster{i});
end

% 调用kmeans
pic = imread('E:\pythonfile_withpycharm\cluster.jpg');
pic = imresize(pic,0.5);
[m,n,z]=size(pic);
row_data = double(reshape(pic,m*n,z));
%---------------------gpu--------------------------%
% 产生一组随机索引
rand_index= randperm(m*n,10);
kcenteroids = row_data(rand_index,:);
max_iter =20;
tic
[clusters, target] = calculateDistforGpu(row_data, kcenteroids, max_iter);
fprintf('kmeans run time：%f\n',toc);
delete(gcp);

% 显示聚类后的图片
img = zeros(m*n,z);
colormap=[0,255,0;255,0,0;0,0,255;200,200,100;180,255,100;
          20,187,240;100,100,150;140,150,160;177,190,0;200,0,200];
for i=1:m*n
    img(i,:)=colormap(target(i),:);
end
img = reshape(img,m,n,z);
res = im2uint8(mat2gray(img));
imshow(res);

原图像为:

聚类之后结果显示: