Using Singular value decomposition for feature extraction from images

Suppose I have a single subject who has 10 images under different conditions. I need to apply singular value decomposition on all 10 images for feature extraction. Then storing the common extracted feature from 10 images.

The purpose is to save space. Rather than saving all 10 images I would like to save common extracted in one place and use the common extracted feature for future face recognition of the same subject (person).

The 10 images have been attached.

[Christine Tobler]

Do you want to compute the singular value decomposition of each of these images and store those? Or compute a combined decomposition of all 10 images?

[Ron Herman]

SVD on each image for feature extraction.

I also want to know feature extraction using SVD

Answer 1

[Christine Tobler]

Image compression using SVD is a pretty common example (although not the most efficient way to compress an image), here are some relevant links:

https://www.youtube.com/watch?v=mSYgi0KCLv8

https://www.mathworks.com/matlabcentral/fileexchange/69916-image-compression-using-svd

https://stackoverflow.com/questions/13614886/using-svd-to-compress-an-image-in-matlab

For feature extraction, I don't know of a "standard" algorithm to do this using SVD. I found some papers searching for this topic, maybe one of these could be helpful:

https://scholar.google.com/scholar?q=feature+extraction+using+svd

Answer 2

[Yee Mon]

% Load the image

image1 = imread('yme.jpg');

% Convert the image to grayscale if it's RGB

if size(image, 3) == 3

image = rgb2gray(image);

end

% Perform Singular Value Decomposition (SVD) on the image matrix

[U, S, V] = svd(double(image));

% Choose the number of singular values to keep (feature extraction)

num_features = 50;

U = U(:, 1:num_features);

S = S(1:num_features, 1:num_features);

V = V(:, 1:num_features);

% Reconstruct the image using the selected features

reconstructed_image = unit8(U * S * V');

% Display the original and reconstructed images by SVD features

subplot(1, 2, 1);

imshow(image1);

title('Original Image');

subplot(1, 2, 2);

imshow(reconstructed_image);

title('Reconstructed Feature Image with SVD');