hey folks, i have a question on FFT filtering that i am working on. My goal is to separate out the periodic part and aperiodic part of a image. This idea comes from Michal Haindl and Martin Hatka's paper'Near-Regular Texture Synthesis'. And it is shown below in the FFT filter part.
I understand that the FFT spectrum of a image, and the low freq. part is most likely to be the periodic part. Since for noise, it is most likely to be in high freq.
For the above image. even though it is not perfectly periodic, but if we do the zeroing and then do the ifft, we should be able to have it doing the job for more or less, this is what I am thinking for now.
What I have done for so far it to fft2 the image and in order to get the spectrum, I had fftshift, and F=abs(F), F=log(F+1). And then I think I have three methods that I can try in order to do the zeroing: 1st, set a threshold and have everything below it to be zeroed. 2nd, manually zeroing some of the spikes(which i know PhotoShop can do in quick), and 3rd, applying some filter on it(which i will try in future if the 1st method can give some promising result).
I am trying the first one to firstly see how would the picture look like after some zeroing, and if it works out, I will go for the filtering. But the problem I am having is:
after doing all the fftshift(F) and abs(F) and F=log(F+1), the F that we have is not capable to do the IFFT. is it the right way to go back if I have F=exp(F)-1; img3=abs(fftshift(ifft2(F))); I get confused cuz what's coming out does not seem to be right direction. I know this will not work, but it should at least give back something that make sense I thought.
Also I was thinking is there any particular filters that you guys would think that is likely to work here.
Please let me know am I on the right track, any help would be very much appreciated
Best Answer