图像分割的阈值算法matlab实现

1、图像分割的阈值算法matlab实现【OTSU,1DEntropy,2DEntropy】今天看了几篇论文，实现了一下，没有验证各算法的有效性也没有进行定量比较OTSU% OTSU method% 2006/9/4clc;clear;%I = imread('E:testchinalake.bmp','bmp');I = imread('E:testlena.png','png');I = double(I);I = Medianfilter(I);       &

2、#160; % median filterh_Tmean = mean(mean(I);height,width = size(I);Size = height * width;  % the size of the imageh_T = sum(sum(I);      % the total gray value of the imageG_min = min(min(I);    % the min gray value of the imageG_max = max(max(I); &#

3、160;  % the max gray value of the iamgeI_seg = zeros(height,width);     % the array to store the segmented imagethresh = 0;          % the thresholdnum1 = 0;num2 = 0;          

4、0; % count the num of the pixel from the diffrient classP1 = 0;P2 = 0;              % the probability of the different classh_T1 = 0;h_T2 = 0;            % the total gray value of

5、different class h_T1mean = 0;h_T2mean = 0;        % the mean value of the classmax = 0;for thresh=G_min:G_max     % find the best threshold    h_T1 = 0;    h_T2 = 0;    num1 = 0;    for

6、 h=1:height        for w=1:width             if I(h,w) <= thresh                  num1 = num1 + 1;   &

7、#160;              h_T1 = h_T1 + I(h,w);             end         end     end     num2 = Size -

8、num1;     h_T2 = h_T - h_T1;     P1 = num1/Size;     P2 = num2/Size;          h_T1mean = h_T1/num1;     h_T2mean = h_T2/num2;          %D =

9、P1*(h_T1mean - h_Tmean)2 + P2*(h_T2mean - h_Tmean)2;     D1 = P1*P2*(h_T1mean - h_T2mean)2;                             % the tow equation i

10、s equal     if D1 > max         max = D1;         T_best = thresh;            % T record the best thresh     end end&

11、#160; % Seg the image % for i=1:height     for j=1:width         if I(i,j) > T_best             I_seg(i,j) = 255;         end&#

12、160;    end end T_best figure; imshow(uint8(I_seg); figure; imhist(uint8(I);      *    一维直方图熵阈值算法% 1D entropy thresholding method% Pun提出，Kapur对其阈值和熵进行改进%  两类：object 和background% P1 = sum(pi)  i：1T% P2

13、 = sum(pi)  i:T+1255% HO = ln(P1)  + H1/P1;% HB = ln(P2) + H2/P2;% H1 = -sum(pi*ln(pi); i:1T% H2 = -sum(pi*ln(pi); i:T+1255% H = HO + HB;% T_best = argmax(H);clc;clear;%I = imread('E:testchinalake.bmp','bmp');I = imread('E:testlena.png','png');I = double(I);

14、I = Medianfilter(I);         % median filterheight,width = size(I);Size = height * width;  % the size of the imageh_T = sum(sum(I);      % the total gray value of the imageG_min = min(min(I);    % the min gray value

15、 of the imageG_max = max(max(I);    % the max gray value of the iamgeI_seg = zeros(height,width);     % the array to store the segmented imageI_hist = zeros(1,256);           % the array to store the hist of the ima

16、gethresh = 0;          % the thresholdnum1 = 0;num2 = 0;            % count the num of the pixel from the diffrient classP1 = 0;P2 = 0;           

17、;   % the probability of the different classh_T1 = 0;h_T2 = 0;            % the total gray value of different class max = 0;H1 = 0;H2 = 0;             % the middle varH_object

18、 = 0;H_background = 0;H_total = 0;        % the total entropyT_best = 0;         % the best thresh% 计算直方图 %for i=1:height        % calculate the hist of the image    for j=1:w

19、idth        I_hist(I(i,j)+1) = I_hist(I(i,j)+1) + 1;    endendfor thresh=G_min:G_max     % find the best threshold    H1 = 0;    h_T1 = 0;    H2 = 0;    for h=1:height

20、60;       for w=1:width             if I(h,w) <= thresh                  num1 = num1 + 1;     

21、60;            h_T1 = h_T1 + I(h,w);             end         end     end     num2 = Size - num1; 

22、60;   h_T2 = h_T - h_T1;     P1 = num1/Size;     P2 = num2/Size;     for i=1:thresh         px = I_hist(i+1)/Size;         H1 = H1 + (-px*ln(px); 

23、;    end     for i=thresh+1:G_max         px = I_hist(i+1)/Size;         H2 = H2 + (-px*ln(px);     end          H_object = l

24、n(P1) + H1/P1;     H_background = ln(P2) + H2/P2;          H_total = H_object + H_background;          if H_total > max         max = H_total;  

25、60;      T_best = thresh;     end end  % Seg the image % for i=1:height     for j=1:width         if I(i,j) > T_best          &

26、#160;  I_seg(i,j) = 255;         end     end end T_best figure; imshow(uint8(I_seg); figure; imhist(uint8(I);*2维直方图熵阈值算法% 二维直方图熵阈值法% 参考 基于2D 熵阈值的铁谱磨粒图像分割方法，傅建平%廖振强，张培林，汪传忠，（南京理工大学机械学院，南京 ），%（军械工程学院，石家庄）%&#

27、160;  pixel gray%   %   |%   |              => 2D histgram%   |%   |%   |_> local grayclc;clear;%I = imread('E:testchinalake.bmp','bmp');I = imread(

28、'E:testlena.png','png');I = double(I);height,width = size(I);Size = height * width;  % the size of the imageG_min = min(min(I);    % the min gray value of the imageG_max = max(max(I);    % the max gray value of the iamgeI_2Dhist = zeros(G_max+1,G_ma

29、x+1);   % the array to store the 2D hist of the imageI_mean = zeros(height,width);   % the mean value of the local imageI_seg = zeros(height,width);WS = 3;                     

30、;     % mean filter's window size 3*3nr = floor(WS/2);I_big = zeros(height+2*nr,width+2*nr);    % the bigger array used to mean filterI_big(nr+1:height+nr,nr+1:width+nr) = I;  % copy data from the original image% mean filter % 获取局部区域灰度信息 %for i=1:height

31、60;   for j=1:width         sum = 0;         num = 0;         for h=-nr:nr             for w=-nr:nr  

32、;               sum = sum + I_big(i+h,j+w);                 num = num + 1;             end

33、         end         I_mean(i,j) = sum/num;     end end  % 构建2D直方图，横轴上以点象素灰度表示，纵轴上以局部区域灰度表示 % for i=1:height     for j=1:width     

34、0;   h = I(i,j)+1;             % 横轴信息,避免0,所以加1，象素灰度         w = I_mean(i,j)+1;        % 纵轴信息，避免0，所以加1，局部区域灰度        

35、; I_2Dhist(h,w) = I_2Dhist(h,w) + 1;  % 统计灰度对<pixel,local>的出现次数，构建2D直方图     end end   % find the best thresh : hor_thresh,and ver_thresh % for ver_thresh=0:G_max     for hor_thresh=0:G_max      

36、    sum1 = 0;          sum2 = 0;          H1 = 0;          H2 = 0;         for i=0:ver_thresh   

37、;          for j=0:hor_thresh                 sum1 = sum1 + I_2Dhist(i+1,j+1);             end   

38、60;     end         for i=0:ver_thresh             for j=0:hor_thresh                 P1 = I_2Dhist(

39、i+1,j+1)/sum1;                 H1 = H1 + P1*log(P1);             end         end      

40、0;          if i < G_max & j < G_max            for i=ver_thresh+1:G_max                 for j=hor_thresh+1:G_max&#

41、160;                   sum2 = sum2 + I_2Dhist(i+1,j+1);                end          &

42、#160; end            for i=ver_thresh+1:G_max                for j=hor_thresh+1:G_max                    P2 = I_2Dhist(i+1,j+1)/sum2;                    H2 = H2 +P2*log(P2);