Make ifft2 to give real output

13 次查看(过去 30 天)
I'm trying to generate a randomly rough surface(isotropic surface). Suppose I have the absolute values of fourier components of the surface in "H" and H has symmetry with respect to center of the array. Now, I need to define some random phase for the fourier series which are between -pi to pi (due to fftshift) as in:
phi = -pi + (pi+pi)*rand(n); % random phase
Now, I don't know what to do to make the output of my surface real..
Here is the gist of the code I used so far:
[a,b] = pol2cart(phi,H);
H_complex = complex(a,b); % the complex fourier transform composed of H & phi
z = ifft2(ifftshift((H_complex)));
If I need to apply conjugate symmetry to the fourier components, how should I do it? The output is complex and I need a real surface.
Please help! Thanks a lot.

采纳的回答

Mona Mahboob Kanafi
编辑:Mona Mahboob Kanafi 2015-5-21
I found my answer, so maybe useful for someone else as well. I need to apply conjugate symmetry condition for 2D fourier transform, both in magnitude values and phase components. I applied it as below for a n*m matrix:
"H" is the magnitude:
H(1,1) = 0;
H(1,m/2+1) = 0;
H(n/2+1,m/2+1) = 0;
H(n/2+1,1) = 0;
H(2:end,2:m/2) = rot90(H(2:end,m/2+2:end),2);
H(1,2:m/2) = rot90(H(1,m/2+2:end),2);
H(n/2+2:end,1) = rot90(H(2:n/2,1),2);
H(n/2+2:end,m/2+1) = rot90(H(2:n/2,m/2+1),2);
Similar operation but with negative sign applies to "phi" (phase):
phi(1,1) = 0;
phi(1,m/2+1) = 0;
phi(n/2+1,m/2+1) = 0;
phi(n/2+1,1) = 0;
phi(2:end,2:m/2) = -rot90(phi(2:end,m/2+2:end),2);
phi(1,2:m/2) = -rot90(phi(1,m/2+2:end),2);
phi(n/2+2:end,1) = -rot90(phi(2:n/2,1),2);
phi(n/2+2:end,m/2+1) = -rot90(phi(2:n/2,m/2+1),2);
You can refer to this link for a good example of 2D fourier transform on a matrix and how the results should look like: http://fourier.eng.hmc.edu/e101/lectures/image_processing/node6.html
  1 个评论
Ralf Mackenbach
Ralf Mackenbach 2019-4-10
Small correction to this. The condition that
H(1,1) = 0;
H(1,m/2+1) = 0;
H(n/2+1,m/2+1) = 0;
H(n/2+1,1) = 0;
Does not need to hold necessarily. So a better (examplary) version would be
% Choose even m and n
m = 100 ;
n = 100 ;
% Choose average value of surface
z0 = 1 ;
% Fill random examplary matrices
H = rand(n,m) ;
phi = 2*pi*rand(n,m) ;
% Apply symmetry to matrices
H(1,1) = z0*m*n ;
H(2:end,2:m/2) = rot90(H(2:end,m/2+2:end),2);
H(1,2:m/2) = rot90(H(1,m/2+2:end),2);
H(n/2+2:end,1) = rot90(H(2:n/2,1),2);
H(n/2+2:end,m/2+1) = rot90(H(2:n/2,m/2+1),2);
phi(1,1) = 0;
phi(1,m/2+1) = 0;
phi(n/2+1,m/2+1) = 0;
phi(n/2+1,1) = 0;
phi(2:end,2:m/2) = -rot90(phi(2:end,m/2+2:end),2);
phi(1,2:m/2) = -rot90(phi(1,m/2+2:end),2);
phi(n/2+2:end,1) = -rot90(phi(2:n/2,1),2);
phi(n/2+2:end,m/2+1)= -rot90(phi(2:n/2,m/2+1),2);
% Convert to polar
[a,b] = pol2cart(phi,H);
H_complex = complex(a,b);
% Inverse and plot
S = ifft2(H_complex, 'symmetric') ;
surf(S)

请先登录,再进行评论。

更多回答(1 个)

Thomas Koelen
Thomas Koelen 2015-5-20
use abs() or real()? depending on if you want only the real number or the absolute value.
  1 个评论
Mona Mahboob Kanafi
编辑:Mona Mahboob Kanafi 2015-5-20
Thanks Thomas, I need the output to be real based on a solid theory. Since, from the beginning, I have to define the absolute values of the fft so that it gives a surface with specified standard deviation (root mean square roughness). Real values, don't have the same std and absolute values have the same std, but not a correct shape! Absolute values are only positive while my surface must have a zero mean. This is not an option for me at the moment.

请先登录,再进行评论。

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by