Prev: datenum and data problem
Next: max
From: Royi Avital on 15 Mar 2010 03:25
If I may add, have you encountered anything better than Wiener Filter.
I'm haveing troubles with it as well.

I can easily extract the PSF (Or very good estimation of it) yet suffer from Ringing.
I think Maximum Likelihood algorithm would be much better.

I'm in search of a modern working Maximum Likelihood Algorithm (Something better than "deconvblind").

If anyone knows something about it, it would be great.
From: James caron on 15 Mar 2010 10:36
"Royi Avital" <RoyiREMOVEAvital(a)yahoo.com> wrote in message <hnknch$iba$1(a)fred.mathworks.com>...
> If I may add, have you encountered anything better than Wiener Filter.
> I'm haveing troubles with it as well.
>
> I can easily extract the PSF (Or very good estimation of it) yet suffer from Ringing.
> I think Maximum Likelihood algorithm would be much better.
>
> I'm in search of a modern working Maximum Likelihood Algorithm (Something better than "deconvblind").
>
> If anyone knows something about it, it would be great.

I have had nothing but great success using a Wiener filter for deconvolution. It receives a bad reputation because it is often not applied correctly. Three things to look at:
1) If the artifacts occur at all edges in the image, then either your PSF is inaccurate or it is not clean. If there is noise in the PSF image, it will turn into artifacts in the restoration. If the PSF is not sufficient, then you should go to blind deconvolution (I recommend SeDDaRA).
2) If the ringing is from the edges, then you should 'reflect' the image out such that in frequency space it is continuous. This slows processing a bit, but is much better than apodization.

Hope that helps,
Jim C
From: Royi Avital on 16 Mar 2010 14:34
"James caron" <na(a)matlab.com> wrote in message <hnlgla$iec$1(a)fred.mathworks.com>...
> "Royi Avital" <RoyiREMOVEAvital(a)yahoo.com> wrote in message <hnknch$iba$1(a)fred.mathworks.com>...
> > If I may add, have you encountered anything better than Wiener Filter.
> > I'm haveing troubles with it as well.
> >
> > I can easily extract the PSF (Or very good estimation of it) yet suffer from Ringing.
> > I think Maximum Likelihood algorithm would be much better.
> >
> > I'm in search of a modern working Maximum Likelihood Algorithm (Something better than "deconvblind").
> >
> > If anyone knows something about it, it would be great.
>
> I have had nothing but great success using a Wiener filter for deconvolution. It receives a bad reputation because it is often not applied correctly. Three things to look at:
> 1) If the artifacts occur at all edges in the image, then either your PSF is inaccurate or it is not clean. If there is noise in the PSF image, it will turn into artifacts in the restoration. If the PSF is not sufficient, then you should go to blind deconvolution (I recommend SeDDaRA).
> 2) If the ringing is from the edges, then you should 'reflect' the image out such that in frequency space it is continuous. This slows processing a bit, but is much better than apodization.
>
> Hope that helps,
> Jim C

Could you elaborate about "Smoothing" it in the frequency domain?
I guess I'm after the best Iterative algorithm once you have a great initialization of the PSF estimated by my other Algorithm.
From: James caron on 17 Mar 2010 10:47

> > I have had nothing but great success using a Wiener filter for deconvolution. It receives a bad reputation because it is often not applied correctly. Three things to look at:
> > 1) If the artifacts occur at all edges in the image, then either your PSF is inaccurate or it is not clean. If there is noise in the PSF image, it will turn into artifacts in the restoration. If the PSF is not sufficient, then you should go to blind deconvolution (I recommend SeDDaRA).
> > 2) If the ringing is from the edges, then you should 'reflect' the image out such that in frequency space it is continuous. This slows processing a bit, but is much better than apodization.
> >
> > Hope that helps,
> > Jim C
>
> Could you elaborate about "Smoothing" it in the frequency domain?
> I guess I'm after the best Iterative algorithm once you have a great initialization of the PSF estimated by my other Algorithm.

Assuming you use an image of a point source as your PSF, the FFT of that image will have the spatial frequencies associated with the actual PSF plus all the noise frequencies. By applying a smoothing filter to the FFT of the PSF, you smooth the noise in the frequency domain which reduces the amplification of the noise during the deconvolution without affecting the actual PSF.

Jim C
 | 
Pages: 1
Prev: datenum and data problem
Next: max