pca2d.m

用matlab实现了2dpca算法

clear all;

t0=clock;
class_num = 40;
 class_sample = 10;
 train_num = 5;
 test_num = 5;
 scale = 1;
path = ['D:\tool\FACE-ORL\ORL-'];
 d = 8;
for i=1:class_num
    for j =1:train_num
        name =[path num2str(i,'%.3d') '\' num2str(j,'%.2d') '.BMP' ];
        [a,map]=imread(name,'bmp');
        a = imresize(a,scale);
        Train_Image (:,:,(i-1)*train_num+j) = a;
    end;
end;
ImageSize = size(a);
Train_Image = double(Train_Image);
Cov = zeros(ImageSize(2),ImageSize(2));
mean_Image = mean(Train_Image,3);
Train_ImageSub = zeros(ImageSize(1),ImageSize(2),class_num*train_num);
for i=1:class_num*train_num
    Train_ImageSub(:,:,i) = Train_Image(:,:,i) - mean_Image(:,:);

    Cov = Cov + Train_ImageSub(:,:,i)'*Train_ImageSub(:,:,i);
end;
Cov = Cov/(class_num*train_num);

[V,D] = eig(Cov);
d1 = diag(D);
[d2,index] = dsort(d1);
v_sort = zeros(ImageSize(2),d);
for i=1:d
    d_sort(i) = d1(index(i));
    v_sort(:,i) = V(:,index(i));
    v_sort(:,i) = v_sort(:,i)./sqrt(sum(v_sort(:,i).^2));
end;

for i=1:class_num*train_num
    Coe(:,:,i) = Train_Image(:,:,i)*v_sort;
end;

accu = 0;
for i=1:class_num
    for j =train_num+1:class_sample
        name =[path num2str(i,'%.3d') '\' num2str(j,'%.2d') '.BMP' ];
        Image = double(imread(name));
        Image = imresize(Image,scale);
        CoeTest= Image*v_sort;
        for k=1:class_num*train_num
               dist(k) = sum(sum((CoeTest - Coe(:,:,k)).^2)); 
        end;
           [min_d,min_index] = min(dist);

         if floor((min_index-1)/train_num)==(i-1)
                    accu = accu+1;
          end;
    end;
end;
 accu/(class_num*test_num)
 time1 = etime(clock,t0)