deconvolution vs unsharp mask

[bird (Anthony Wesley)]

I was asked the other day how deconvolution is different to unsharp masking... they both are used to "sharpen" an image by removing blur.

I don't really know the mathematics of deconvolution, but I have an example here that shows it in action, and also shows how powerful it can be.

My saturn imaging from last week was plagued with horizontal line noise, maybe my LU camera is faulty or maybe the power supply for it was picking up something nasty, I don't know. But after stacking and wavelet processing in registax the channels were looking pretty bad. Unsharp masking would have exaggerated the problem so I couldn't use that approach... so I tried deconvolution instead.

Here is a screen shot showing the three channels with the registax image at the top and the result of deconvolution processing, and the final recombined image. The application is Astra Image. Notice how deconvolution removes the horizontal noise and *at the same time* it sharpens the real signal.

This screenshot is about 160kb in size.

http://www.acquerra.com.au/personal/...15/montage.jpg

regards, Bird

[RB (Andrew)]

That's so cool, thanks Anthony for the info.

I was mucking around with Astra Image on my solar spot image after reading your post and I couldn't believe the difference it made using deconvultion instead of unsharp.

I'm seriously thinking of buying a copy.

Thanks again.

[Robert_T]

Thanks Bird you're a Marvel

I've got the demo copy of Astra Image so I'll have to try it out. A lot of my images have been degraded by lines and hatching that are exacerbated by unsharp masking so maybe this will help.

cheers,

[janoskiss (Steve H)]

If you see anything that says "maximum entropy" in your image processing software, use it!

[rumples riot]

Thanks for the demonstration Anthony. Nice Saturn incidently.

[Robert_T]

Hi All - finally loaded up the demo version of Astra Image 2.0 and immediately bought and registered it. This is the program I've been looking for and I've been able to put deconvolution to the test for the first time. As with Bird's original post where unsgarp masking would have also sharpened up lines noise etc. Using LR deconvolution (haven't played much with ME yet) managed to sharpen up some of my better images without exacerbating the slight bayer hatching present. See the two images attached - at left after wavelet processing in Reg 3 and small amt unsharp masking in Pixel Studio 2.0, and at right after splitting the colour panes and running LR deconvolution (5 iterations and curve=2) and recombining in Astra Image 2.0. There is a slight "ring" effect around the limb (anyone know how to get rid of this?), but the detail is much clearer

cheers,

[bird (Anthony Wesley)]

A lot comes down to picking good parameters for deconvolution - I usually try a few different sets of values and compare them carefully to see which ones are better. The default psf radius is 2.0 in AA, but i normally find that this is too large and end up with something like 1.5, or even someitmes close to 1.0.

Maximum entropy is another one to try, it gives quite different results, and also Van Cittert. Other items to play with FFT filtering.

I really like the way that AA creates a separate window for each result, it's a nice interface.

Bird

[Robert_T]

Thanks Bird, I'll give the parameters a tweak and try ME as you suggested - think they're the only options (?) in the standard Astra Image (as opposed to Astra Image Max? Now I don't even need the sky to clear, can just do astraimage re-runs of last months best