Re: [APML] Re: Van Cittert Deconvolution

["Mike Unsold" <mike@mlunsold.com>]

Juan,

Your experimental VC method is an improvement. Some of the fine noise
amplification
still remains. Are you applying the same set of band-pass filters to each
pixel at each
iteration?

The classical Richardson-Lucy algorithm does suffer from ringing and speckle
noise.
Removing these defects is the goal of the improved RL algorithm. The
improved
version gives much better control over ringing and noise and thus allows the
application
of a lot more iterations before they start to appear.

Another technique that can be used to reduce the artifacts produced by a
restoration
methods is to blend different restored versions of the same image. For
example,
60 iterations could restore the image but produce ringing around the bright
regions.
20 iteration could partially restore the image without ringing. The maximum
combination
of the 60 and 20 iteration images would have the highly restored bright
regions of the
60 iteration image blended with the smooth no ring dark regions of the 20
iteration image.

In the link below the left image is the result of 60 iterations, the middle
image had 20 iterations,
and the right image is the maximum combination of the left and middle image.

http://www.mlunsold.com/examples/Temp/FranksPlato2.jpg

Currently I have implementations of the Van Cittert, One-Step Gradient,
classical and adaptive
Richardson-Lucy, and maximum entropy restoration methods. I am interested in
extending
each of the methods using wavelets to determine significant structure in an
image. An image's
significant structure could then be used to guide the iterative restoration
process.

Mike


----- Original Message -----
From: "Juan Conejero" <skycad@ctv.es>
To: "Discussion of Film Astrophotography" <astro-photo@seds.org>
Sent: Saturday, March 15, 2003 10:10 PM
Subject: [APML] Re: Van Cittert Deconvolution


> Hi Mike,
>
> Very good results with that improved Richardson-Lucy.
>
> We are currently working on a modification to the classical Van Cittert's
> algorithm which greatly avoids noise amplification. Basically, we are
> applying the idea of suppressing high-frequency layers of the image at
each
> iteration. This can be accomplished through wavelets, but we are having
> quite good results with combinations of band-pass filters, which are much
> faster. Here is our try on the Frank Barret's image with our experimental
> algorithm:
>
>
http://www.pleiades-astrophoto.com/software/experiments/deconvolution/etc/Pl
ato-VC.jpg
>
> Only ten iterations of our frequency-limited Van Cittert's algorithm. The
> PSF was a gaussian function of radius 2.0.
> We like the Van Cittert's method because its simplicity and fast
operation.
> Richardson-Lucy is more prone to artifacts (at least in our experience),
> like ringing.
>
> A similar mechanism of frequency layer suppression will be included in the
> next version of SGBNR.
>
> Regards,
> ______________________________________
> Juan Conejero, Pleiades Astrophoto
> skycad@ctv.es
> http://www.pleiades-astrophoto.com/en.html
>
> At 18:00 15/03/2003 -0500, you wrote:
> >Hi Frank,
> >
> >The noise in the dark regions of your lunar image is from noise
> >amplification
> >by the Van Cittert deconvolution. This is a drawback with the Van Cittert
> >method.
> >Other methods have been devised to mitigate noise amplification. The
> >following
> >image is the result of 40 iterations of an adaptive Richardson-Lucy
> >deconvolution
> >using a 7x7 Gaussian point spread function. This method has much better
> >resistance
> >to noise amplification.
> >
> >http://www.mlunsold.com/examples/Temp/PlatoCompare.jpg
> >
> >Mike
>
>
> _______________________________________________
> Astro-Photo mailing list
> Astro-Photo@seds.org
> http://seds.org/mailman/listinfo/astro-photo
>


_______________________________________________
Astro-Photo mailing list
Astro-Photo@seds.org
http://seds.org/mailman/listinfo/astro-photo