When I am taking the derivative of the minimization criterion with respect to X(k,l), I can’t find the expression you give next.
Shouldn’t the minimization criterion be writen:
sum(|Y(k,l)-H(k,l)X(k,l)|²+alpha|L(k,l)X(k,l)|²)?
Maybe I have misunderstood.
And don’t you write at the numerator of the filter H*(k,l)?

Thank you for your help!

I am interested in the application of Maximum Entropy deconvolution, is there a Maximum Entropy deconvolution function available in Matlabs image processing toolbox ??

I have a question though, i am using this to try and deconvolve free space propagation x-ray phase contrast profiles. When i supply a profile as input ,in class double, which is in the range between say 0.8 and 1.2 and roughly centred on 1 my resulting profile is not in this range. The deconvolution looks good but the returned values are between 0.025 and 0.006 say and centred at around 0.013, do you have any idea waht might be causing this? Everything is in double precision when input into deconvreg.

Please let me know if i can provide any other info and i look forward to your reply.

Thanks in advance,

Tom H

This and the Weiner filtering have been very useful to me.

I’m currently working on an Inverse Halftone project and trying to estimate images based on error-diffused halftones.

I’m able to use the deblur techniques and toolboxes you’ve mentioned in these two blogs with some success – from a distance the regularization image does appear fairly similar to the original image, but it’s requiring me to use a fairly high alpha (about .9), so I’m losing clarity on high-frequency sections of the image, particularly with color images. Have you done much with inverse halftoning and do you have any suggestions? Thanks much!

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

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.