Deconvolution problems for stars against nebula and background.

[Lord Beowulf]

Rather than copying everything here, I'll point people to this thread on CNfor the detail and pictures of what I'm encountering.  Basically if I adjust the decon settings so that the ringing looks good against the nebula, things look horrible in the background.  If I make the background levels look good, the nebula has dark rings around all the stars.  Any input on a clean "PI" way to fix this (vs. do it twice and merge the two with masking) I'd appreciate the feedback.

Thanks,

Beo

[ngc1535]

Rather than copying everything here, I'll point people to this thread on CNfor the detail and pictures of what I'm encountering.  Basically if I adjust the decon settings so that the ringing looks good against the nebula, things look horrible in the background.  If I make the background levels look good, the nebula has dark rings around all the stars.  Any input on a clean "PI" way to fix this (vs. do it twice and merge the two with masking) I'd appreciate the feedback.

Thanks,

Beo

I waited to see if there were going to be replies but haven't seen any yet... so I will jump in and offer my observations for what they are worth.

1. The general interplay you are looking is a balance between dark ringing seen against a bright background and a not objectionable "glow/halo" for stars against the dark sky background by manipulating the global dark deringing. Based on your images, you might be leaning more in one direction. Where your stars are against the dark background and you show a halo- you also have a darkness beyond that point. This indicates your mask might need an adjustment in terms of both size and fuzziness.
2. I am particularly confused by the affect on the background surrounding your stars. It appears that the mask is quite large (too large) surrounding stars in general and your global dark setting is too high. The masking of stars (basically bright signal) usually needs to be just as large as the extent that compensation of the ringing is necessary. No larger than this spatial scale.
3. The local support you are using may or may not be helpful here. It is hard to tell without seeing the interplay of your mask and your local support structure map. This kind of question really needs the inventory of support images (mask and local support) to be visible/available.
4. In my experience with oversampled images (such as yours) I find the LR decon result converges (with strong sharpening) after only 20-30 iterations. 100-500 iterations seems like quite a bit to me (but take that for what it is worth... I understand deconvolution is a blind algorithm with no guarantee of convergence). The 20-30 iterations is value I demonstrate in my instructional stuff.
5. I am assuming your image is a 16-bit linear critter. Something that is 15-bits might be harder since more bright stars will not be available for decon.

I hope something in the above is helpful to you.
-adam

[Lord Beowulf]

Thanks for the response Adam.  I received a response on CN last night that had me try something different that gave me a much better result.  I was using the Luminance (CIE Y) target on my RGB image for decon for those initial results.  I went back and tried extracting the luminance to its own channel and just running on that and got VERY different results.  With no deringing local support, I was able to just push the global dark setting up from 0.06 to 0.40 and pretty much eliminate the rings.  I was also able to push the number of iterations up considerably (although I'll need to pull it back slightly I expect) from the 10 iterations I originally had up to 30.  After that I pieced it back together with the extract RGB information to get back to a color image.  Not sure why PI can't work with the luminance directly in the RGB image, but for some reason it's obviously very different.

Thanks,

Beo