Preliminary test with Astra Image deconvolution

[michael.h.f.wilkinson]

Just a side by side comparison

The first image is processed in Registax 6, with my usual wavelet scheme, the second in Astra Image Plus 4.0. The latter with Lucy Richardson deconvolution 1.5 pixel Gaussian kernel and some 100 iterations with strength of 1.7 or so. I applied a gamma of 0.75 as well. The deconvolved image is clearly sharper, but the contrast of the finer detail is less strong (more natural, really)

[Luke]

Cor,  number two looks a clear winner to my eyes, Michael. Nice one!

[Stardust]

you wouldn't think it was the same image, will have to look at this Astra Image. So are you doing the stack in Registax then taking it to AI for the processing?

[Steve Ward]

I would say that having a "usual sharpening scheme" is a big mistake , even for my humble "white" stuff I find that no two days data is alike (today's being an extreme example ...  ) and I very rarely have the same wavelets settings from one day to the next .

Another great deconv program is "DStation" , as used by Harald Palaske on his mindblowing solar close-ups , extremely tunable an multi-optioned.

https://github.com/blackhaz/DStation

I would also suggest that you knock the bottom three or four Wavelet sliders over to the far left and use only the top two small radius sliders , with a gentle touch , to bring out the finer detail , the bottom large radius controls ruin many of the images I see posted across various sites .

Edited October 15, 2014 by Steve Ward

[Merlin66]

AstroArtV5 also does the Richardson Lucy. (as well as Van Cittert, Wiener etc.)

[michael.h.f.wilkinson]

I would say that having a "usual sharpening scheme" is a big mistake , even for my humble "white" stuff I find that no two days data is alike (today's being an extreme example ...  ) and I very rarely have the same wavelets settings from one day to the next .

Another great deconv program is "DStation" , as used by Harald Palaske on his mindblowing solar close-ups , extremely tunable an multi-optioned.

https://github.com/blackhaz/DStation

I would also suggest that you knock the bottom three or four Wavelet sliders over to the far left and use only the top two small radius sliders , with a gentle touch , to bring out the finer detail , the bottom large radius controls ruin many of the images I see posted across various sites .

I wasn't being clear. I typically use a base scheme as starting point, and tweak that until it is (visually) optimal. I do notice I rarely deviate very much from the base settings I stored. That is what I meant by usual wavelet scheme. I actually find my few white-light images require more experimentation than my H-alpha.

[michael.h.f.wilkinson]

you wouldn't think it was the same image, will have to look at this Astra Image. So are you doing the stack in Registax then taking it to AI for the processing?

I tend to stack in AS!2. I then tended to import into Registax 6. That will be AI now, I think.

[michael.h.f.wilkinson]

A little unsharp masking in GIMP  improves the result:

Wavelets (used curves to match tonal range)

Original LR deconvolution

LR deconvolution with unsharp masking (default settings)

LR deconvolution with unsharp masking (more aggressive)

The latter is perhaps a bit too strong. Incidentally, the image processing toolbox in MatLab does support LR deconvolution, so I might well see if I can write a script to do the deconvolution of all the panes

[JB80]

Right, another good comparison.

I have had a fiddle this morning with 3.0 and am I right in thinking you can't save it in the free version?

Might have to bite the bullet.

edit: Just found you have to use the license Key.

Edited October 16, 2014 by JB80

[michael.h.f.wilkinson]

Right, another good comparison.

I have had a fiddle this morning with 3.0 and am I right in thinking you can't save it in the free version?

Might have to bite the bullet.

edit: Just found you have to use the license Key.

You do need a licensed version in order to save the images. I got the 4.0 Plus variant. Quite cheap really.

[JB80]

I thinks it's too early in the day for me, I was wondering why I had the key but hadn't been asked for it. Sorted now.

It is quite cheap so I'm playing around with 3.0 to see if I will go ahead and get 4.0, stick with 3.0 or PS. It's a close call so far.

[Zakalwe]

Bliley....every one is going Astra Image crazy!

Personally I haven't seen that much improvement on disc detail compared to multiple iterations of Unsharp Mask (about 10-15 consecutive repetitions of very gentle UM). It really comes into its own on faint proms though

[michael.h.f.wilkinson]

Unsharp mask does not really do a deconvolution, which is what you really want to do to reduce the effect of seeing etc. Unsharp mask is effectively a local contrast stretch.

[Montana]

Excellent comparisons :)

I think the main point is that wavelets are just too harsh for solar processing and not a good method. The Sun does not have hard edges, it is a soft ball of flowing plasma and so usual wavelets have a hard time and tend to bring out hard edges where there is none. Using a mild deconvolution method or Smart sharpen is a very gentle way and will just bring focus without creating false hard lines or halos. I don't think it matters which you use, everyone has different equipment and different cameras have different levels of noise. Just be gentle, less is more, go for the real fluffy look, not the hard sharp edges look

Alexandra

[michael.h.f.wilkinson]

I think I will be tinkering away at at least one of my mosaics to reprocess all images, and stitch them anew. I might have a shot at using Hugin instead of AutoStitch64.

[Pete Presland]

interesting comparison and thread.