Eigenface Face Recognition: Why do I keep getting "Subscripted assignment dimension mismatch"?

I was working on a code that I got from https://blog.cordiner.net/2010/12/02/eigenfaces-face-recognition-matlab/. I was able to make it work. Then after taking a break, I came back to the code and started getting "Subscripted assignment dimension mismatch."

Here's the code:

%%Loading the Images
clear all
close all
clc
input_dir = 'C:\Users\Nadlei\Desktop\dlsufacereg\Training';
image_dims = [30, 64];

filenames = dir(fullfile(input_dir, '*.jpg'));
num_images = 30;
images = zeros(prod(image_dims),num_images);

for n = 1:num_images
filename = fullfile(input_dir, filenames(n).name);
img = imread(filename);
%img = imresize(img,[30,64]);
img = im2double(img);
images(:, n) = img(:);
end

%%Training
% steps 1 and 2: find the mean image and the mean-shifted input images
mean_face = mean(images, 2);
shifted_images = images - repmat(mean_face, 1, num_images);

% steps 3 and 4: calculate the ordered eigenvectors and eigenvalues
[evectors, score, evalues] = pca(images');

% step 5: only retain the top ‘num_eigenfaces’ eigenvectors (i.e. the principal components)
num_eigenfaces = 20;
evectors = evectors(:, 1:num_eigenfaces);

% step 6: project the images into the subspace to generate the feature vectors
features = evectors'*shifted_images;

% calculate the similarity of the input to each training image
input_image = imread('input.jpg');
input_image = imresize(input_image,image_dims);
input_image = im2double(input_image);
feature_vec = evectors' * (input_image(:) - mean_face);

similarity_score = arrayfun(@(n) 1 / (1 + norm(features(:,n) - feature_vec)), 1:num_images);

% find the image with the highest similarity
[match_score, match_ix] = min(similarity_score);

% display the result
figure, imshow([input_image reshape(images(:,match_ix), image_dims)]);
title(sprintf('matches %s, score %f', filenames(match_ix).name, match_score));

The error occurs at: "images(:,n) = img(:);" I know that it means that the matrix assignments don't match, but how do I fix it? All my input images are already in grayscale and resized to Width:30 and Height:64. Someone please help me.

Answer 1

[Walter Roberson]

One of your images is not the same size as the others. You have an imresize() there that would take care of that problem, but you have that imresize() commented out.

[Jess Marcos]

Unfortunately, even after adding that line, I still get the error. I modified it to:

images(:, n) = str2double(img(:));

But then I get an "Index exceeds matrix dimensions" for:

evectors = evectors(:, 1:num_eigenfaces);

[Jess Marcos]

This was the code that was working up until last night.

%%Loading the Images
input_dir = 'C:\Users\Nadlei\Desktop\dlsufacereg\Training\';
image_dims = [64, 30];
filenames = dir(fullfile(input_dir, '*.jpg'));
num_images = numel(filenames);
images = zeros(prod(image_dims), num_images);
for n = 1:num_images
    filename = fullfile(input_dir, filenames(n).name);
    img = imread(filename);
    img = im2double(imresize(img, image_dims));
    images(:, n) = img(:);
end
%%Training
% steps 1 and 2: find the mean image and the mean-shifted input images
mean_face = mean(images, 2);
shifted_images = images - repmat(mean_face, 1, num_images);
% steps 3 and 4: calculate the ordered eigenvectors and eigenvalues
[evectors, score, evalues] = pca(images');
% step 5: only retain the top ‘num_eigenfaces’ eigenvectors (i.e. the principal components)
num_eigenfaces = 30;
evectors = evectors(1:num_eigenfaces);
% step 6: project the images into the subspace to generate the feature vectors
features = evectors'*shifted_images;
% calculate the similarity of the input to each training image
input_image = imread('input.jpg');
input_image = imresize(input_image,image_dims);
input_image = im2double(input_image);
feature_vec = evectors' * (input_image(:) - mean_face);
similarity_score = arrayfun(@(n) 1 / (1 + norm(features(:,n) - feature_vec)), 1:num_images);
% find the image with the highest similarity
[match_score, match_ix] = min(similarity_score);
% display the result
figure, imshow([input_image reshape(images(:,match_ix), image_dims)]);
title(sprintf('matches %s, score %f', filenames(match_ix).name, match_score));

[Walter Roberson]

Change

for n = 1:num_images
    filename = fullfile(input_dir, filenames(n).name);
    img = imread(filename);
    img = im2double(imresize(img, image_dims));
    images(:, n) = img(:);
end

to

for n = 1:num_images
    filename = fullfile(input_dir, filenames(n).name);
    [img, map] = imread(filename);
    if isempty(img)
      error('Somehow file "%s" has an empty image!', filename);
    end
    if ~isempty(map)
      error('Somehow file "%s" is a pseudocolor JPEG image, which is not possible in standard JPEG!', filename);
    end
    if ndims(img) > 2
      img = zeros(image_dims);
      warning('File "%s" was an RGB JPEG image (as nearly all JPEG images are), but the code expects it to one of the rare grayscale JPEG images. Substituting black image instead.', filename);
    else
      img = im2double(imresize(img, image_dims));
    end
    images(:, n) = img(:);
end

[Jess Marcos]

The next error I get is: "Index exceeds matrix dimensions." at:

evectors = evectors(:, 1:num_eigenfaces);

[Walter Roberson]

The only possibility that comes to mind is if you only had a single image then you would be sending a row vector to pca and perhaps it treats row vectors differently?

Ah, there is another explanation: your num_images might be 0, so your images variable might be empty.

Put in a breakpoint and look at the size of images and evectors and let us know what you found.

Answer 2

[Neelesh Patel]

in same code i have error occured in that line;

feature_vec = evectors' * (input_image(:) - mean_face);
**Matrix dimensions must agree.**

[Walter Roberson]

input_image is not the same size as the images you trained on.

[Neelesh Patel]

hello sir i m trying to execute this code:

%%Loading the Images
input_dir ='E:\TARGET FOLDER';
image_dims = [64, 30];
filenames = dir(fullfile(input_dir, '*.jpg'));
num_images = numel(filenames);
images = zeros(prod(image_dims), num_images);
for n = 1:num_images
    filename = fullfile(input_dir, filenames(n).name);
    [img, map] = imread(filename);
    if isempty(img)
      error('Somehow file "%s" has an empty image!', filename);
    end
    if ~isempty(map)
      error('Somehow file "%s" is a pseudocolor JPEG image, which is not possible in standard JPEG!', filename);
    end
    if ndims(img) > 2
      img = zeros(image_dims);
%       warning('File "%s" was an RGB JPEG image (as nearly all JPEG images are), but the code expects it to one of the rare grayscale JPEG images. Substituting black image instead.', filename);
    else
      img = im2double(imresize(img, image_dims));
      img = rgb2gray(img)
    end
    images(:, n) = img(:);
end
%%Training
% steps 1 and 2: find the mean image and the mean-shifted input images
mean_face = mean(images, 2);
shifted_images = images - repmat(mean_face, 1, num_images);

% steps 3 and 4: calculate the ordered eigenvectors and eigenvalues
[evectors, score, evalues] = pca(images');

% step 5: only retain the top ‘num_eigenfaces’ eigenvectors (i.e. the principal components)
num_eigenfaces = 20;
evectors = evectors(:, 1:num_eigenfaces);
% step 6: project the images into the subspace to generate the feature vectors
features = evectors'*shifted_images;
% calculate the similarity of the input to each training image
input_image = imread('E:\TARGET FOLDER\0.jpg');
input_image = imresize(input_image,[64, 30]);
input_image = im2double(input_image);
input_image = rgb2gray(input_image);
input_image = size(input_image);
feature_vec = evectors' * (input_image(:) - mean_face);

i m getting error in last step of this code

plzzz plzz help to solve it out....thanx sir in advance

[Walter Roberson]

What are you expecting the line

input_image = size(input_image);

to do for you?

[Neelesh Patel]

Because of this error i have used that size function plzz sir solve it out because i m hunger for the result......

feature_vec = evectors' * (input_image(:) - mean_face); Error using * Inner matrix dimensions must agree.

[Neelesh Patel]

else give me another code which gives me specific result because its a project of 'POLICE' in india so i have to design it with full effort so plz help sir....................

[Walter Roberson]

"because its a project of 'POLICE' in india"

I am not able to assist with police or military related projects in any country. (Not even my own -- my security clearance lapsed about a month ago.)

You need to assume that anyone without a security clearance for your country, who provides you with code for facial recognition for police or military use, is a foreign or criminal antagonist with an interest in making the project fail in hidden ways.

You need to re-read the documentation on size() and you need to learn how to use the debugger to solve this problem for yourself.

Answer 3

[Neelesh Patel]

Hello sir Plzz help me for this error:

feature_vec = evectors' * (input_image(:) - mean_face); Error using - Matrix dimensions must agree.

[Walter Roberson]

You need to re-read the documentation on size() and you need to learn how to use the debugger to solve this problem for yourself.