Using Singular value decomposition for feature extraction from images
조회 수: 8 (최근 30일)
이전 댓글 표시
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.
댓글 수: 2
Christine Tobler
2020년 4월 17일
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?
답변 (2개)
Christine Tobler
2020년 4월 17일
Image compression using SVD is a pretty common example (although not the most efficient way to compress an image), here are some relevant links:
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:
댓글 수: 0
Yee Mon
2024년 3월 13일
% 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');
댓글 수: 0
참고 항목
카테고리
Help Center 및 File Exchange에서 Eigenvalues에 대해 자세히 알아보기
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!