Siril 1.2.0, Windows 11
I haven’t seen this before so not sure if it’s a bug. Deconvolution works as expected if I allow it to generate the PSF using Blind Deconvolution. If I use Dynamic PSF to select the stars, and then tell Deconvolution to use ‘PSF from Star’, I get what looks like selected stars or ringing that’s color instead of the expected black ringing if settings were to high after applying. I am removing the stars from Dynamic PSF after the PSF Preview is generated, but before clicking Apply. Using deconvolution with the same settings in each case.
Hi Rich, no quick answer I’m afraid. I can replicate your result - it appears to occur with RGB images when the PSF is substantially bigger than it should be. So there are a couple of aspects to this: 1. why is the PSF generated from star fitting substantially bigger than that worked out by blind deconvolution? and 2. why the color shift when applying the final deconvolution?
On the first point, at least on my images that I’ve been investigating with, I get a considerably closer fit to the blind deconvolution PSF if I change the star fitting type from Gaussian to Moffat. I haven’t experimented exhaustively, but it might be that that’s generally a better star model to use when preparing for deconvolution.
On the second point, I’m not sure yet. The deconvolution algorithm transforms the RGB into XYZ color space and only deconvolves the Y (linear intensity) channel, then recombines with the original X and Z channels. I can’t see anything obviously wrong in the code, but I still have some work to do there.
For the time being I’d recommend sticking with the blind deconvolution or estimating your PSF using Moffat star profiles, but hopefully I’ll have a more satisfactory answer soon.
Thanks for the response and I’ll use the blind as suggested. Just so you’re aware I did select Moffat in the Dynamic PSF, and also a range of 0.1-0.7. Thanks again.
I’ve just committed a fix to 1.2. Using a RGB to YUV transform works much better: rather annoyingly I’m not 100% sure why, because I’ve checked the RGB to XYZ transform several times and it’s definitely correct. Anyway. I’ve just pushed a fix to the 1.2 branch, so you can either download the artefact from gitlab once the pipeline has finished running or it will be in the 1.2.1 release.
1 Like