Comments on: Image deblurring using regularization http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/ Steve Eddins manages the Image & Geospatial development team at <a href="http://www.mathworks.com/">The MathWorks</a> and coauthored <a href="http://www.mathworks.com/support/books/book5291.html?category=-1&language=-1">Digital Image Processing Using MATLAB</a>. He writes here about image processing concepts, algorithm implementations, and MATLAB.<br><br><img> Mon, 23 Nov 2009 00:11:16 +0000 http://wordpress.org/?v=2.3.1 By: Steve http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-21943 Steve Mon, 06 Jul 2009 15:20:14 +0000 http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-21943 Shilpa—Using a constant, K, is simply a useful approximation for Sn/Sf, since often you don't really know either Sn or Sf. K can be adjusted until you get a reasonable result. Shilpa—Using a constant, K, is simply a useful approximation for Sn/Sf, since often you don’t really know either Sn or Sf. K can be adjusted until you get a reasonable result.

]]> By: Shilpa http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-21909 Shilpa Wed, 24 Jun 2009 12:33:52 +0000 http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-21909 Dear sir, I have tried to deblur images using Wiener filter. first, I simulated motion blur making use of the degradation function H as given in Book "Digital Image Processing by Gonzalez and Woods". After Evaluating H, the FFT of undegraded image is multiplied with H to obtain blurred image. Gaussian noise with zero mean is then added to this blurred image using MATLAB function imnoise. The degraded image is then deblurred with Wiener filter equation. The (Sn/Sf) ratio is calculated as (Sn/Sf) = ((abs(FFT2(noise)))^2)/(((abs(FFT2(undegraded image)))^2) [ref: DIPUM: GOnzalez, Woods and Eddins]. This ratio (Sn/Sf) is obtained as matrix. My question is, whether this ratio obatined in form of matrix is correct or is it to be converted to some constant K (Some Authors have used a constant K, instead).if it is required to do so, how to convert (Sn/ Sf) to a constant K? I am getting satisfactory results using ratio (Sn/Sf) as matrix. Further, what is the relationship between variance and SNR (dB) for M-by-N noise matrix generated using matlab function imnoise? Thank you Shilpa Dear sir,
I have tried to deblur images using Wiener filter. first, I simulated motion blur making use of the degradation function H as given in Book “Digital Image Processing by Gonzalez and Woods”.

After Evaluating H, the FFT of undegraded image is multiplied with H to obtain blurred image. Gaussian noise with zero mean is then added to this blurred image using MATLAB function imnoise.

The degraded image is then deblurred with Wiener filter equation. The (Sn/Sf) ratio is calculated as
(Sn/Sf) = ((abs(FFT2(noise)))^2)/(((abs(FFT2(undegraded
image)))^2)
[ref: DIPUM: GOnzalez, Woods and Eddins].

This ratio (Sn/Sf) is obtained as matrix.

My question is, whether this ratio obatined in form of matrix is correct or is it to be converted to some constant K (Some Authors have used a constant K, instead).if it is required to do so, how to convert (Sn/ Sf) to a constant K? I am getting satisfactory results using ratio (Sn/Sf) as matrix.

Further, what is the relationship between variance and SNR (dB) for M-by-N noise matrix generated using matlab function imnoise?

Thank you
Shilpa

]]> By: Xin http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-20920 Xin Tue, 29 Jul 2008 09:23:51 +0000 http://blogs.mathworks.com/steve/2008/07/21/image-deblurring-using-regularization/#comment-20920 Regularization is good to perform deblurring or reconstruction. Specified regularization matrix can be used to preserve certain information of the image processed. Now I am using regularization to do a reconstruction. It is similar to deblurring. I have finished its simulation. But as I apply it on real data, a problem appears. In my problem, A*x=b. * refer to the convolution. Using numerical analysis, I convert it to Hy=c. That is, I use multiplication to replace the convolution. So for Hy = c, I can use regularization to obtain its reasonable solution. However, when the convolution kernel is bigger than what we often use, H matrix will be huge and even to the degree that I can not render it in my computer. Moreover, the matrix is not sparse. Do you know there is any methods to deal with the huge matrix H? Thanks. Regularization is good to perform deblurring or reconstruction. Specified regularization matrix can be used to preserve certain information of the image processed. Now I am using regularization to do a reconstruction. It is similar to deblurring. I have finished its simulation. But as I apply it on real data, a problem appears. In my problem, A*x=b. * refer to the convolution. Using numerical analysis, I convert it to Hy=c. That is, I use multiplication to replace the convolution. So for Hy = c, I can use regularization to obtain its reasonable solution.

However, when the convolution kernel is bigger than what we often use, H matrix will be huge and even to the degree that I can not render it in my computer. Moreover, the matrix is not sparse.

Do you know there is any methods to deal with the huge matrix H?

Thanks.

]]>