The use of ifftshift on 2D image analysis

Hello all,

I have a question about ifftshift. It's not a very hard one I'm just not sure I understood what I need from it correctly.

I have an image and I want to take its fourier transform (fft2) and rescale its elements according to a formula which depends on the location (m,n) of its components. To make the (0,0) elements go to the center I first use FFT shift to put it into its standard form.

Now if I want to use ifft2 to go back should I use ifftshift before or after ifft2? If I don't use ifftshift I get a result from ifft2 that also looks correct, but with differently pixel values.

Any help would be appreciated.

Thanks, Mark.

FYI: The equation I want to use is in the following paper, I don't think it's relevant but feel free to take a look if you want. I apologise if you don't have access: http://www.sciencedirect.com/science/article/pii/S0030401896004543

Answer 1

[Walter Roberson]

If your sequence is y = fftshift(fft2(x)) then you should ifft2(ifftshift(y))