BPCA_L1_Reconstruction.rar_1范数_BPCA_L1范数_l1

%%提取所有图像主成分，计算各个主成分数量下的所有图像的重构误差的均值，画出平均重构误差—主成分数量图 clc clear close tic path(path,'D:\MATLAB\function\') Num_subject=15; %样本人数 Num_image=11; %每人图片数 Total_num=165; %总图片数 Fig_row=100; Fig_col=100; % Fig_row=100; % Fig_col=80; % Fig_row=64; % Fig_col=64; % Fig_row=32; % Fig_col=32; Subfig_row=Fig_row/10;%对于64*64和32*32图片，10变为8和4 Subfig_col=Fig_col/10;%同上 Subfig_pixl=Subfig_row*Subfig_col;%每个子图的像素点数 Subfig_num=Fig_row*Fig_col/(Subfig_row*Subfig_col);%每张图分成的子图数量 Train_num=8; %每人训练样本数 Test_num=3; %每人测试样本数量 Noise_minrow=15; %最低噪声行数 Noise_mincol=10; %最低噪声列数 Noise_num=4; %每人训练样本中带噪声的图片数量 PCA_num=100; %需要显示的主成分个数1-min(Subfig_pixl,Subfig_num*Total_num) Cycle_num=1; %每个主成分下，循环Cycle_num次计算重构误差，每次循环均重新加载数据 %% for k=1:Cycle_num if Fig_col==100 load(['D:\MATLAB\YaleFace\165mat\' 'Yale_100x100.mat']); elseif Fig_col==80 load(['D:\MATLAB\YaleFace\165mat\' 'Yale_100x80.mat']); elseif Fig_col==64 load(['D:\MATLAB\YaleFace\165mat\' 'Yale_64x64.mat']); elseif Fig_col==32 load(['D:\MATLAB\YaleFace\165mat\' 'Yale_32x32.mat']); end X_fig=zeros(Fig_row,Fig_col,Total_num); %将图片向量随机存储到X_fig中，其中人的顺序不变 for i=1:Num_subject temp=randperm(Num_image); for j=1:Num_image X_fig(:,:,(i-1)*11+j)=reshape(fea((i-1)*11+temp(j),:),Fig_row,Fig_col); end end %存储原始图片并标准化 X_org=[];%X_org表示不带噪声的图像数据 for i=1:Total_num X_org = [X_org , Fig2Fig(X_fig(:,:,i),Subfig_row, Subfig_col,Subfig_row*Subfig_col, 1, 1,Subfig_num )]; end X_mean=sum(X_org,2)/(Subfig_num*Total_num); X_org=X_org-repmat(X_mean,1,Subfig_num*Total_num);%影响不大 %每人（前Train_num幅）中的随机Noise_num幅加入随机大小（不小于Noise_minrow*Noise_mincol）随机位置的噪声 for i=1:Num_subject %产生带噪声的图像的序号（1-Train_num中任意Noise_num幅图像） temp=randperm(Train_num); temp=temp(1:Noise_num); for j=1:Noise_num X_fig(:,:,temp(j)+11*(i-1))=Matrix_Noise(X_fig(:,:,temp(j)+11*(i-1)),Noise_minrow,Noise_mincol,255); end end %存储加噪图片并标准化 X=[];%X表示带噪声的图像数据 for i=1:Total_num X = [X , Fig2Fig(X_fig(:,:,i),Subfig_row,Subfig_col,Subfig_row*Subfig_col , 1, 1 , Subfig_num )]; end X_mean_N=sum(X_org,2)/(Subfig_num*Total_num); X=X-repmat(X_mean_N,1,Subfig_num*Total_num);%影响不大 %采用PCA-L1提取主成分 U=PCA_L1(X,PCA_num); %计算PCA_num从1—PCA_num时的各个图片的重构误差 X_org=X_org+repmat(X_mean,1,Subfig_num*Total_num);%如果X_org上面经过标准化，则此处应还原回去 X=X+repmat(X_mean_N,1,Subfig_num*Total_num);%如果X上面经过标准化，则此处应还原回去 e=zeros(PCA_num,Cycle_num);%重构误差，一维表示1-PCA_num个PCA情况，二维表示Cycle_num次循环情况 for i=1:PCA_num for j=1:Total_num %计算所有图片的重构误差的均值 temp=X_org(:,((j-1)*Subfig_num+1):(j*Subfig_num));%取一幅原图像 e_temp=zeros(1,Subfig_num); for q=1:Subfig_num e_temp(1,q)=norm(temp(:,q)-U(:,1:i)*(U(:,1:i)'*temp(:,q)),2)^2; end e(i,k)=e(i,k)+sqrt(sum(e_temp))/Total_num; end end toc disp(['循环次数：',num2str(k),'/',num2str(Cycle_num )]) tic end %显示平均重构误差—主成分数量图 figure(1) hold on grid on plot(1:PCA_num,sum(e,2)/Cycle_num,'b') save D:\MATLAB\BPCA_L1_Reconstruction\BPCA_L1_Reconstruction(PCA_100_Cyc_1).mat saveas(gcf,['D:\MATLAB\BPCA_L1_Reconstruction\','BPCA_L1_Reconstruction'],'fig') %显示所有原图、加噪原图、2DPCA_L1重构图 for i=1:Total_num imwrite(mat2gray(Fig2Fig(X_org( : ,((i-1)*Subfig_num+1):i*Subfig_num) , 100 ,1 , 10 , 10 , 10 , 10 )) ,['D:\MATLAB\BPCA_L1_Reconstruction\Output_Figure\Total_Original_Figure\',num2str(i),'.bmp']) imwrite(mat2gray(Fig2Fig(X( : , ((i-1)*Subfig_num+1):i*Subfig_num) , 100 ,1 , 10 , 10 , 10 , 10 )) ,['D:\MATLAB\BPCA_L1_Reconstruction\Output_Figure\Total_Noise_Figure\',num2str(i),'.bmp']) imwrite(mat2gray(Fig2Fig(U*U'*X( : , ((i-1)*Subfig_num+1):i*Subfig_num) , 100 ,1 , 10 , 10 , 10 , 10 )) ,['D:\MATLAB\BPCA_L1_Reconstruction\Output_Figure\Total_Rconstruction_Figure\',num2str(i),'.bmp']) end toc