CFA Drizzle Integration

[multiweb]

Hi there, I'm having some issues running a drizzle integration with subs shot with an OSC (QHY8).
My flow is to bring already calibrated frames into PI as FITS file. From there I do a VNG/Multi Resolution Support (RGGB) debayer to color.
Then a star alignment, Generate drizzle data checked, all default kept but for Bicubic B-Spline Pixel interpolation.
First image integration is Average combine, all defaults, Generate drizzle data checked, Pixel Rejection (1) Winsorized Sigma Clipping.
Final Drizzle Integration Scale 2, all default (Drop shrink 0.9).
If I tick Enable CFA drizzle I get the lines and moire effect (bottom half of attachment).
If I leave Enable CFA drizzle unticked, the resolution is not as good but I don't get any artefacts in the final.

Original files are undersampled. FSQ106N 530mm FL with QHY8 OSC 7.8um pixels so image scale of 3.04 second of arc per pixel. Subs were 300s x 23.
PI Version 01.08.05.1353 Ripley (x64) Win7

Thanks for any tips.

[John_Gill]

Hi,

I think the "Enable CFA drizzle" should be left unchecked because you have already debayered the images earlier in your workflow.  The debayer process converts the color image to a monochrome image.

Hope this helps.
Look up
John

[ajniskan]

Hi there, I'm having some issues running a drizzle integration with subs shot with an OSC (QHY8).
My flow is to bring already calibrated frames into PI as FITS file. From there I do a VNG/Multi Resolution Support (RGGB) debayer to color.
Then a star alignment, Generate drizzle data checked, all default kept but for Bicubic B-Spline Pixel interpolation.
First image integration is Average combine, all defaults, Generate drizzle data checked, Pixel Rejection (1) Winsorized Sigma Clipping.
Final Drizzle Integration Scale 2, all default (Drop shrink 0.9).
If I tick Enable CFA drizzle I get the lines and moire effect (bottom half of attachment).
If I leave Enable CFA drizzle unticked, the resolution is not as good but I don't get any artefacts in the final.

Original files are undersampled. FSQ106N 530mm FL with QHY8 OSC 7.8um pixels so image scale of 3.04 second of arc per pixel. Subs were 300s x 23.

Hi!

I'm using CFA drizzle processing flow quite frequently, but I've not seen a result like in your image ever. The sequence of pre-processing steps looks right, though. Here are some thoughts: 23 frames for CFA drizzle seems to me a bit on the low side. Have you checked you have enough movemed between images to successfully do the CFA drizzle? Did you use guiding when you shot the frames and if you did, how frequently did you dither? And as a troubleshooting step, did you try to do the CFA drizzle step with 1.0 scale?

kind regards,
Antti

[multiweb]

Hi John, the CFA drizzle works with the information collected during the debayering process and the initial star alignment saved in the DRZ files according to the documentation. Debayering will give you a color file interpolated back to native resolution, so slight blurring, from a mono CFA. I assume it works as advertised. If you look closely at the stellar profiles , they are much tighter and defined when CFA Drizzle is ticked.

Thanks Antti, that's the answer I was kind of expecting, not that I like hearing the confirmation I didn't dither these subs and the guiding was unfortunately good enough from frame to frame to have a significant overlap. I'll keep this in mind for my next data acquisition and factor in large amount of dithering. Regarding the number of subs I though 20+ was a respectable number. How many subs do you routinely stack? I'm portable, not observatory based and the imaging sessions are far in between for me so I tend to do only a couple of hours on the same target.

Final question which is related to the cosmetic process. My aging CCD has a lot of hot/dead pixels. I can get rid of them by subtracting a dark or a bad pixel map. The resulting CFA file has "punched holes" in the bayer matrix. If it was a mono I would interpolate the "zero" pixels with neighbouring but if I do that with a CFA you end up generating color noise in one channel as you're mixing one colour with adjacents colors. How do you "fill in" those missing pixels with the cosmetic process? There doesn't seem to be much settings available to tweak. Or is there a better way?

[ajniskan]

Thanks Antti, that's the answer I was kind of expecting, not that I like hearing the confirmation I didn't dither these subs and the guiding was unfortunately good enough from frame to frame to have a significant overlap. I'll keep this in mind for my next data acquisition and factor in large amount of dithering. Regarding the number of subs I though 20+ was a respectable number. How many subs do you routinely stack? I'm portable, not observatory based and the imaging sessions are far in between for me so I tend to do only a couple of hours on the same target.

Final question which is related to the cosmetic process. My aging CCD has a lot of hot/dead pixels. I can get rid of them by subtracting a dark or a bad pixel map. The resulting CFA file has "punched holes" in the bayer matrix. If it was a mono I would interpolate the "zero" pixels with neighbouring but if I do that with a CFA you end up generating color noise in one channel as you're mixing one colour with adjacents colors. How do you "fill in" those missing pixels with the cosmetic process? There doesn't seem to be much settings available to tweak. Or is there a better way?

Hi Marc,

my most successful CFA drizzles have been with 60+ subs. This leads me to use shorter exposures possibly with no guiding at all. Dithering obviously takes time, so I've been experimenting with compromising just slightly on my pole alignment. This limits my exposure times to less than 120 s, but with 60 s I get very nice stars and natural dithering over the session. I've ended up doing a target re-alignment after it has drifted more than 5 % from the original location. It's less scientific than I'd like, though

Regarding the dead pixels, have you checked out the SplitCFA script (https://pixinsight.com/forum/index.php?topic=6249.0)? If I understand the functioning (and what you were hoping to do) correctly, you should be able to interpolate more safely with the CFA channels separated. Then, you'd just apply MergeCFA at some point. The option for CFA drizzle would remain, if you just do the pixel interpolation and go back to merged CFA for later steps of your post process.

kind regards,
Antti

[multiweb]

Thanks Antti, didn't think about doing it this way. I'll have a look into it. At a quick glance I've just noticed the MergeCFA process works only with views so I figure it wouldn't work with an Image container in batch mode.