ICA.rar_ICA_ICA 识别_ica代码_ica程序代码

clear; select=1; switch select case 1 [IMG M N c p NN]=ORL2();%c=40总人数;p=10每人10幅图像;NN=M*N图像大小;A读取图像矩阵 case 2 [IMG M N c p NN]=FERET2();%c=200总人数;p=7每人7幅图像;NN=M*N图像大小;A读取图像矩阵 case 3 [IMG M N c p NN]=Yale2();%c=15总人数;p=11每人11幅图像;NN=M*N图像大小;A读取图像矩阵 case 4 [IMG M N c p NN]=AR2();%c=120总人数;p=26每人26幅图像;NN=M*N图像大小;A读取图像矩阵 case 5 [IMG M N c p NN]=face_db2(); end ln=5; %训练样本数 train=zeros(1,ln); %存储训练样本位置 test=zeros(1,(p-ln)); %存储测试样本位置 %[train test]=RandSelect(ln,p);%从每人p幅图像中随机抽取ln个训练样本，剩余p-ln幅作测试样本 %选择训练样本 selectsample=0; switch selectsample case 0 train=1:ln; %采用前面ln幅图像训练样本 test=ln+1:p; %剩余p-ln幅图像作测试样本 case 1 %随机选择ln幅作训练样本 train=[2]; test=Gettestsample(p,train); case 2 train=[3,5]; test=Gettestsample(p,train); case 3 train=[1,3,12,]; test=Gettestsample(p,train); case 4 train=[5,10,12,18]; test=Gettestsample(p,train); case 5 train=[3,9,17,25,20]; test=Gettestsample(p,train); case 6 train=[2,5,8,11,18,19]; test=Gettestsample(p,train); case 7 train=[2,7,10,15,18,22,23]; test=Gettestsample(p,train); case 8 train=[4,9,10,18,13,19,21,24]; test=Gettestsample(p,train); case 9 train=[3,9,11,13,17,19,20,22,23]; test=Gettestsample(p,train); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%ICA%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% fprintf('************ICA实验结果***********************\n'); c1=clock; X1_=zeros(NN,ln*c); for i=1:c for j=1:ln X1_(:,j+(i-1)*ln)=IMG(:,i,train(j)); end end %X2_是测试样本集 X2_=zeros(NN,(p-ln)*c); for i=1:c for j=1:p-ln X2_(:,j+(i-1)*(p-ln))=IMG(:,i,test(j)); end end %将训练样本中心化 mx=zeros(NN,1); for i=1:ln*c mx=mx+X1_(:,i); end mx=(1/(ln*c)).*mx; X1=zeros(NN,ln*c); for i=1:ln*c X1(:,i)=X1_(:,i)-mx; end %白化及求解混阵 x=X1'; k=70;%选择投影轴数 [P1 R]=prewhitening_1(x,k);%P1为相互正交矩阵，而R代表一对角矩阵 P2=(inv(R)*P1'*x)'; Z=P2'*X1;%球化 ini=randn(k,k);%随即产生初始一个初始权矢量（随机的W，使其维数为Z的行向量个数） epsilon=0.01; g=1; Wd=FastICA(ini,Z,epsilon,g); W=Wd*P2'; %%%%%%%进行识别%% feature1=W*X1_; %抽取训练样本特征 feature2=W*X2_; %抽取测试样本特征 c2=clock; fprintf('特征抽取时间:');etime(c2,c1) %特征抽取时间 wrong=0; for i=1:c for j=1:(p-ln) d=1e6; a2=feature2(:,j+(i-1)*(p-ln)); for a=1:c for b=1:ln a1=feature1(:,b+(a-1)*ln); d0=a2-a1; d0=norm(d0); if d>d0 d=d0; index=a; end end end if index~=i; wrong=wrong+1; end end end c3=clock; fprintf('分类时间:');etime(c3,c2) %分类时间 fprintf('总识别时间:');etime(c3,c1) %总识别时间 fprintf('识别率:');r=(c*(p-ln)-wrong)/(c*(p-ln)) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%