inaccurate result of pca face recognition

hi, can you guys please help me. i don't know where i did wrong. i've use PCA method for face recognition. and for the recognition process i've used euclidean distance. but the result of input test image is different from recognized image.

this coding has 320 training images and 80 test images.

%calculate training feature vector
train_featureVec=[];
for i=1:size(faceRegistered,2)
    train_featureVec(:,i)=normal(:,i)'*pca_project;
end

%load the test image database
for i=1:80
    test_im{i,1}=imread(sprintf('AT&T Test/test_im (%d).jpg',i));
end

faceTest = zeros(112*92,80*1);
%row=total pixel,column=total images

%calculate test feature vectors
weightTest=[];
for j=1:size(faceTest,2)
    %find weight of test image
    weightTest(:,j)=faceTest(:,j)-u1(:,1);
    test_featureVec(:,j)=normal(:,j)'*pca_project;
end

%recognition process
dist_mat =[];
for a=1:80
    input_filename=imread(sprintf('AT&T Test/test_im (%d).jpg',a));
    test_featureVec1=test_featureVec(:,a);
    for b=1:320
        registered_filename=faceRegistered;
        train_featureVec1=train_featureVec(:,b);
        diff_vector=test_featureVec1-train_featureVec1;
        dist_mat(1,b)=sqrt(sum(abs(diff_vector)).^2);
    end

%recognition process
minimum= min(dist_mat);
[r,c]=find(dist_mat==minimum) %find the position of the minimum distance
if size(c)==size([1 2]) % !!!error has been fix here!!!
    c=c(1,2);
    count=c;
else
    count=c; %count=column with minimun distance
end
%classRP=classRegisteredPerson(count,:); % Class of Registered Person

%display input vs recognized face from training set
result_filename = ['AT&T Database/train_im (',num2str(count) ,').jpg'];
result_face_image = imread(result_filename);
 
%Show the input face vs recognized face
figure(3);
subplot(121);    
imshow(input_filename);
title('Input Face');

subplot(122)
imshow(result_face_image);
title('Recognized Face');
end

Answer 1

[aditi bagora]

It appears that the problem lies in how values are assigned to the "dist_mat" matrix. Currently, the indexing used is (1, b), which means that only the distance corresponding to the last test image is being stored in the matrix. To resolve this issue, you can employ the following approach to ensure all distances are properly stored.

dist_mat (a, b) = sqrt(sum(abs(diff_vector)). ^2)

Also, to find minimum distance and column with minimum distance for a test image. You can use the “dist_mat” by indexing.

[minimum, count] = min (dist_mat (i, :)) where i, is index of test image.

Hope the provided approach helps you in getting accurate results.