Wiener filter image restoration

 
    
%parameter T (observation time) and motion rate
T=1; ax=30; ay=40; NSR=0; %since no noise
 
%reading the blurred picture
I=im2double(imread('blur.bmp'));
 
%generating frequencies for the blurring model
u=linspace(-0.5,0.5,size(I,2));
v=linspace(-0.5,0.5,size(I,1));
 
[U,V]=meshgrid(u,v);
H=(T./(pi*(U*ax+V*ay))).*sin(pi*(U*ax+V*ay)).*exp(-1i*pi*(U*ax+V*ay));
 
I_f=fft2(I);
I_motion_fn=fftshift(I_f);
 
wiener=(1./H).*((H.^2)./((H.^2)+NSR));
 
I_recon_fn=I_motion_fn.*wiener;
I_recon=ifft2(ifftshift(I_recon_fn));
 
figure(1);
subplot(1,2,1), imagesc(I), colormap(gray)
title('original image')
 
subplot(1,2,2),imagesc(abs(I_recon))
title('reconstruction using wiener')

Im trying to restore the following image using a wiener filter as show above, however im not getting the reconstructed image, what is the issue? I know that a wiener filter with no noise acts like an ideal inverse filter, however applying the code above is not showing me the reconstructed image

Also, if I were to do the same code but using the built in deconvwnr filter, how would I go around that using the parameters I have?

Answer 1

[Image Analyst]

Try imshow() with [] instead:

subplot(1,2,2); % Display in the right hand axes.
imshow(abs(I_recon), []);

[Kat_33]

I did try that before, it doesnt solve my problem unfortunately

[Image Analyst]

Why do you say you can't get I_recon? Does the fft2() or ifftshift() not work or throw an error or something? It's not just going to finish with no I_recon being created unless an error is thrown. Do you see I_recon in the workspace? Do you see any red error text (that you forgot to attach)? Please reply after you read this link.

Or do you actually get I_recon (contrary to what you said), but you just don't like how it looks?

[Kat_33]

As I said, I'm not getting anything that looks like the reconstructed image, its simply noise without any trace of the original picture

[Image Analyst]

Then I guess you didn't pick the correct inverse filter - the same one used to blur the image in the first place.

[Kat_33]

But I didn't apply any filter to blur the image. It is already blurred and I'm trying to unblur it using the wiener formula as I wrote it above(without using deconvwnr), not sure why its not working

[Image Analyst]

Rather than starting out with some image where you don't know the blur kernel, start out with a known image, and then blur it with a known kernel. Then use that to see how well you can undo the blur using the known kernel.