Right now I am trying to match a small template image to a bigger one to see where the correlation is the highest.
I know that I can compute this by filtering the big image with a filter comprised by the image values of the small template.
However, this is really slow.
I also know that I can circumvent this filtering by transferring the images into the FFT domain, where the convolution becomes an easy pointwise multiplication.
But here I am stuck. Now the images are of different size in the FFT domain, how can I ever multiply them? Or do I need to do a FFT with regard to the same size? And if yes, which one is it?