matlab下利用K-Means进行图像分类
FIRST & BEST SOLUTION
clear all; clc; I_rgb=imread('dog.jpg'); figure();imshow(I_rgb);title('原始图像'); %去噪 filter=ones(5,5); filter=filter/sum(filter(:)); denoised_r=conv2(I_rgb(:,:,1),filter,'same'); denoised_g=conv2(I_rgb(:,:,2),filter,'same'); denoised_b=conv2(I_rgb(:,:,3),filter,'same'); denoised_rgb=cat(3, denoised_r, denoised_g, denoised_b); D_rgb=uint8(denoised_rgb); figure();imshow(D_rgb);title('去噪后图像');%去噪后的结果 %将彩色图像从RGB转化到lab彩色空间 C =makecform('srgb2lab'); %设置转换格式 I_lab= applycform(D_rgb, C); %进行K-mean聚类将图像分割成2个区域 ab =double(I_lab(:,:,2:3)); %取出lab空间的a分量和b分量 nrows= size(ab,1); ncols= size(ab,2); ab =reshape(ab,nrows*ncols,2); nColors= 4; %分割的区域个数为4 [cluster_idx,cluster_center] =kmeans(ab,nColors,'distance','sqEuclidean','Replicates',2); %重复聚类2次 pixel_labels= reshape(cluster_idx,nrows,ncols); %显示分割后的各个区域 segmented_images= cell(1,4); rgb_label= repmat(pixel_labels,[1 1 3]); for k= 1:nColors color = I_rgb; color(rgb_label ~= k) = 0; segmented_images{k} = color; end figure(),imshow(segmented_images{1}),title('分割结果——区域1'); figure(),imshow(segmented_images{2}),title('分割结果——区域2'); figure(),imshow(segmented_images{3}),title('分割结果——区域3'); figure(),imshow(segmented_images{4}),title('分割结果——区域4');%使分割后的图像在一个图中显示 m=uint8(rgb_label); for i=1:69 for j=1:97 if m(i,j,1)==1 m(i,j,1)=255; m(i,j,2)=0; m(i,j,3)=0; end if m(i,j,1)==2 m(i,j,1)=256; m(i,j,2)=256; m(i,j,3)=0; end if m(i,j,1)==3 m(i,j,1)=0; m(i,j,2)=0; m(i,j,3)=255; end if m(i,j,1)==4 m(i,j,1)=0; m(i,j,2)=128; m(i,j,3)=0; end end end figure(),imshow(m)
将 调用k-means算法的那句更换成下面代码，自己实现k-means
cluster_idx=zeros(6693,1); ct11=90; ct12=90; ct21=110; ct22=110; ct31=130; ct32=130; ct41=150; ct42=150; %初始分类 sum1=[0,0]; sum2=[0,0]; sum3=[0,0]; sum4=[0,0];
f1=0;
f2=0;
f3=0;
f4=0;
for k=1:20 for i=1:6693 d1=(ab(i,1)-ct11).^2+(ab(i,2)-ct12).^2; d2=(ab(i,1)-ct21).^2+(ab(i,2)-ct22).^2; d3=(ab(i,1)-ct31).^2+(ab(i,2)-ct32).^2; d4=(ab(i,1)-ct41).^2+(ab(i,2)-ct42).^2;
Z=[d1,d2,d3,d4]; m=min(Z); if m==d1 cluster_idx(i)=1;
f1=1+f1; sum1=sum1+ab(i,:); end if m==d2 cluster_idx(i)=2;
f2=f2+1; sum2=sum2+ab(i,:); end if m==d3 cluster_idx(i)=3;
f3=f3+1; sum3=sum3+ab(i,:); end if m==d4 cluster_idx(i)=4;
f4=f4+1; sum4=sum4+ab(i,:); end end ct11=sum1(1,1)/f1; ct12=sum1(1,2)/f1; ct21=sum2(1,1)/f2; ct22=sum2(1,2)/f2; ct31=sum3(1,1)/f3; ct32=sum3(1,2)/f3; ct41=sum4(1,1)/f4; ct42=sum4(1,2)/f4; end ct1=[ct11,ct12]; ct2=[ct21,ct22]; ct3=[ct31,ct32]; ct4=[ct41,ct42];
ANOTHER SOLUTION
RGB= imread ('dog.jpg'); %读入图像
[m n]=size(RGB); %m是数据个数，n是数据维度
figure(),imshow(RGB);title(' 图一 彩色原图像')
hold off;
RGB=double(RGB);
filter=ones(5,5);
filter=filter/sum(filter(:));
denoised_r=conv2(RGB(:,:,1),filter,'same');
denoised_g=conv2(RGB(:,:,2),filter,'same');
denoised_b=conv2(RGB(:,:,3),filter,'same');
denoised_rgb=cat(3, denoised_r, denoised_g, denoised_b);
RGB=uint8(denoised_rgb);
figure();imshow(RGB);title('去噪后图像');%去噪后的结果
RGB=double(RGB);
img1= RGB(:,:,1);
img2=RGB (:,:,2);
img3= RGB (:,:,3);
t=0;
c11(1)=4; c12(1)=4; c13(1)=4;
c21(1)=70; c22(1)=67; c23(1)=71;
c31(1)=100; c32(1)=100; c33(1)=100;
c41(1)=200; c42(1)=200; c43(1)=200;%选四个初始聚类中心
cluster_idx=zeros(69,97);
class1_num=0;
class2_num=0;
class3_num=0;
class4_num=0;
sum_class11=0;
sum_class21=0;
sum_class31=0;
sum_class41=0;
sum_class12=0;
sum_class22=0;
sum_class32=0;
sum_class42=0;
sum_class13=0;
sum_class23=0;
sum_class33=0;
sum_class43=0;
for k=1:20
if t==0
for i=1:69
for j=1:97
r=sqrt((img1(i,j)-c11(k))^2+(img2(i,j)-c12(k))^2+(img3(i,j)-c13(k))^2);
g=sqrt((img1(i,j)-c21(k))^2+(img2(i,j)-c22(k))^2+(img3(i,j)-c23(k))^2);
b=sqrt((img1(i,j)-c31(k))^2+(img2(i,j)-c32(k))^2+(img3(i,j)-c33(k))^2);
q=sqrt((img1(i,j)-c41(k))^2+(img2(i,j)-c42(k))^2+(img3(i,j)-c43(k))^2); %计算各像素灰度与聚类中心的距离
Z=[r,g,b,q];
d=min(Z);
if d==r
class1_num=class1_num+1;
cluster_idx(i,j)=1;
sum_class11=sum_class11+img1(i,j);
sum_class12=sum_class12+img2(i,j);
sum_class13=sum_class13+img3(i,j);
end
if d==g
class2_num=class2_num+1;
cluster_idx(i,j)=2;
sum_class21=sum_class21+img1(i,j);
sum_class22=sum_class22+img2(i,j);
sum_class23=sum_class23+img3(i,j);
end
if d==b
class3_num=class3_num+1;
cluster_idx(i,j)=3;
sum_class31=sum_class31+img1(i,j);
sum_class32=sum_class32+img2(i,j);
sum_class33=sum_class33+img3(i,j);
end
if d==q
class4_num=class4_num+1;
cluster_idx(i,j)=4;
sum_class41=sum_class41+img1(i,j);
sum_class42=sum_class42+img2(i,j);
sum_class43=sum_class43+img3(i,j);
end
end
end
c11(k+1)=sum_class11/class1_num;
c21(k+1)=sum_class21/class2_num;
c31(k+1)=sum_class31/class3_num;
c41(k+1)=sum_class41/class4_num;%将所有低灰度求和取平均，作为下一个低灰度中心
c12(k+1)=sum_class12/class1_num;
c22(k+1)=sum_class22/class2_num;
c42(k+1)=sum_class42/class4_num;
c32(k+1)=sum_class32/class3_num;%将所有低灰度求和取平均，作为下一个中间灰度中心
c13(k+1)=sum_class13/class1_num;
c23(k+1)=sum_class23/class2_num;
c43(k+1)=sum_class43/class4_num;
c33(k+1)=sum_class33/class3_num;%将所有低灰度求和取平均，作为下一个高灰度中心
d11=abs(c11(k+1)-c11(k));
d12=abs(c12(k+1)-c12(k));
d13=abs(c13(k+1)-c13(k));
d21=abs(c21(k+1)-c21(k));
d22=abs(c22(k+1)-c22(k));
d23=abs(c23(k+1)-c23(k));
d31=abs(c31(k+1)-c31(k));
d32=abs(c32(k+1)-c32(k));
d33=abs(c33(k+1)-c33(k));
d41=abs(c41(k+1)-c41(k));
d42=abs(c42(k+1)-c42(k));
d43=abs(c43(k+1)-c43(k));
if(d11<=0.001&&d12<=0.001&&d13<=0.001&&d21<=0.001&&d22<=0.001&&d23<=0.001&&d31<=0.001&&d32<=0.001&&d33<=0.001&&d41<=0.001&&d42<=0.001&&d43(k)<=0.001)
t=1;
end
end
end
for i=1:69
for j=1:97
if cluster_idx(i,j)==1
img1(i,j)=255;
img2(i,j)=0;
img3(i,j)=0;
end
if cluster_idx(i,j)==2
img1(i,j)=256;
img2(i,j)=256;
img3(i,j)=0;
end
if cluster_idx(i,j)==3
img1(i,j)=0;
img2(i,j)=0;
img3(i,j)=255;
end
if cluster_idx(i,j)==4
img1(i,j)=0;
img2(i,j)=128;
img3(i,j)=0;
end
end
end
Img1=uint8(img1);
Img2=uint8(img2);
Img3=uint8(img3);
R=cat(3,Img1,Img2,Img3);
figure(),imshow(R);title('图二 聚类后的图像')
