mean_shift_iteration.zip_mean_mean shift_mean-shift分割_均值漂移_均值漂移

%% 用向量运算简化时间 ms_dist load('data3003003.mat'); data = double(data); [m,n,z] = size(data); N = m*n; Max=max(max(max(data))); data=data/Max*255; imshow(uint8(data)); data=reshape(data,N,z); %变为二维图像，更好处理 r = 9; zhongxin = data(1,:); array = ms_dist(zhongxin,data,r,m); disp('初始阈值'); epsilon = 0.5; eps2 = epsilon*epsilon; %epsilon是 ε 的意思 % NUM_OF_TRAILS = 20; %随机给定，需要考虑迭代因素 fenge = zeros(N,z); %得到同维数大小的矩阵，存储漂移后的数据 for i =1:N zhongxin = reshape(data(i,:),z,1); %转置为z*1的矩阵 [array] = ms_dist(zhongxin,data,r,m); [mode, score] = mean_shift_iteration(zhongxin, data, array, r, eps2, m); fenge(i,:) = mode; if mod(i,100) == 0 disp('i='),disp(i); end end clear mode; pic_name='漂移2_5.jpg'; fenge1 = reshape(fenge,m,n,z); imwrite(mat2gray(fenge1),strcat(pic_name));%写数据 % % imshow(uint8(fenge)); %此时的fenge只是许多漂移位置的图，不是最终的分割图，通过欧式距离判断归类 fengejieguo = zeros(N,1); r2 = 9; %聚类的容差，应该小于r？？小于r2的归为一类 mode = 1; for i =1:N while fengejieguo(i) == 0 zhongxin0 = reshape(data(i,:),z,1); % data = data(fenge==0,:); %找到未分割的点，减少运算次数 % 不能用，这会导致data顺序变化，得到的weizhi不正确 [array,weizhi] = ms_dist(zhongxin0,data,r,m); zhongxin1 = (sum(array,2)+zhongxin0)/(size(array,2)+1); %取平均值作为新的中心 zhongxin2 = zhongxin1 - r; while ( dist(zhongxin1,zhongxin2) >= r ) [array,weizhi] = ms_dist(zhongxin1,data,r,m); zhongxin2 = zhongxin1; zhongxin1 = sum(array,2) / size(array,2); %sum（a，2）是按行求和 end fengejieguo(i) = mode; fengejieguo(weizhi) = mode; mode = mode + 1; end end disp('模式数目='),disp(mode); jieguo = reshape(fengejieguo,m,n); jieguo = jieguo*250/mode; fengetu = uint8(jieguo); pic_name='漂移结果2_5.jpg'; imwrite(mat2gray(fengetu),strcat(pic_name)); imshow(fengetu);