PCA on an image

Hi there,

I'm trying to apply PCA on an single image and i'am using the same porgram that i used to apply PCA to a grupo of images that are part of the data basis of a neuronal network that i am creating. The thing is that i need to apply PCA to a single image to be that image the input to the neuronal network that i created but it doesnt work.

Anyone has any idea of how can I apply PCA to an image of 50x50 saved as vector of 2500 to have an image of 90?

Thanks a lot.

Answer 1

[Shantanu Dixit]

Hi Laura,

If PCA has already been computed on a group of images, the principal components (eigenvectors) can be reused. The single image can be projected onto these precomputed components to get the reduced representation.

% Assume `coeff` are the eigenvectors precomputed on training images

% Single image (reshape from 50x50 to 1x2500)
image = rand(50, 50); % Example image
image_vector = reshape(image, 1, []);

% Project onto first 90 principal components
reduced_image = image_centered * coeff(:, 1:90);

% Process further to the neural network as required

Hope this helps!

Answer 2

[Image Analyst]

See my attached demo where I compute the principal components of an image.