New Debayer Process/Channel Combination Question

S

Scott Badger

Well-known member
I'm working on IC1396 and playing with the new debayer process to split my OSC (DSLR) subs into individual color channels and I don't understand a difference I'm seeing vs. an integration of the same subs but without splitting the channels. The workflow I follow when not splitting the channels is....
Debayer
SubSelect and weighting
Align
Integrate
Drizzle Integrate
Dynamic Crop

And the result is, as expected, an integrated image with a pervasive green cast that is largely then removed with DBE and the rest via color calibration and SCNR.

On the other hand, using the split color channels option, I....
Debayer into separate color channels
SubSelect and weight the subs for each color channel separately
Align each channel separately, but using the same reference image for all three
Integrate each channel separately
Drizzle integrate each channel separately
Align the three channels (not sure if this is necessary since they were all aligned to the same reference image in the first place)
Dynamic Crop each using the same crop instance
LinearFit the three integrations to the brightest (red in this case)
ChannelCombine the three channels into an RGB

And the resulting integration *doesn't* have the green cast. Why is that?

Thanks,
Scott
 
Hi @Scott Badger,

linear fitting colour channels each other do not make much sense to me. The side effect of doing that is to change the relative ratio between the channels since Linear Fit shifts and scales the signal of each channel to fit the reference one, this means that you're changing the white point of the image and this is the reason why you see that the green cast changes or disappears. This does not mean that this approach produces a correct color calibration, indeed you should use the photometric or manual colour calibration to do that properly (you'll see that the result is different).

So, IMHO you should remove the LineraFit step, just combine the RGB channels and only after do a photometric or manual colour calibration.

Robyx
 
Thanks @robyx ! In posting my question, I was hoping to learn more about LinearFit and when to use it. It's used in a tutorial I reference, but there the data is R, G & B shot separately with a mono camera and filters, not osc images. Does that make a difference, or would you still not use LF to match the color channels? Elephant Trunk is last-light for my DSLR as I just got a 268M, so I'll be shooting R,G&B(&L) separately moving forward.

I'll try dropping the LF step with this image, though when I combine the three channels without LF, the integration then has a bright red cast, so still not the 'normal' green when integrating my rgb images without splitting the channels. Also, fwiw, I wasn't relying on LF to color calibrate. After LF and combining the channels, I would follow up with DBE then color calibration (usually try and compare both manual and photometric). Side note, I've tried applying DBE on the color channels separately before LF and combination, and also after LF and combination, but I don't see much, if any, difference in the result.

Cheers,
Scott
 
If your colour image is from broadband filters (either an OSC camera or mono broadband filters) then the stars in the image are accurate photometric colour references, and provided PCC works (i.e. the image is solvable and there are sufficient APASS catalogue sources in the image) PCC will do an infinitely better job of colour calibration than any other technique. Do not try to do any other LF / color balance process before running PCC.
 
Thanks Fred! Though I usually try both manual and photometric color calibration, I've almost always chosen the photometric version to move forward with. Regarding LF, and just to be clear since it differs from the tutorial, you're saying that if I have integrated images for R, G, and B, that were shot separately with a mono camera, I should *not* use LF to match them before combining them? And is there a point to using LF after combining and color calibrating? Or, maybe better to ask, when is LF appropriate to use?

Cheers,
Scott
 
First, my standard warning: I AM NOT AN EXPERT!
... but as I understand it: before PCC a wide variety of tools were implemented to assist with colour balancing images. Once PCC became available, then if you want photometric colour balance it is really the only tool you need. I (virtually) always run PCC (with background neutralisation) immediately after integration. I have seen suggestions that you might choose to remove strong LP gradients first, but I prefer to run PCC, then remove gradients (typically DBE with loads of manually selected sample points) - I find it easier to judge LP after PCC has got the colour about right. As far as I am aware there is no reason why you shouldn't run PCC again after this (to "re-polish" the colour calibration) but I doubt if it is necessary for normal image processing - which will usually modify the colour balance for aesthetic purposes anyway.
 
Interesting.... I've always used DBE before PCC but I get your point of judging where to put the sample points after correcting the color. As an alternative, I usually pull a lightness image from the RGB and slightly over stretch it to give me a better idea what's background and what's nebulosity, and use that to map the sample points. With Elephant Trunk, finding any true background is a bit tricky.....

This all still brings me back around to my original question, though slightly re-stated.... Why does an integration of the separate channels split out from my osc images show a red cast (when LF isn't applied) instead of the green cast that results from integration without splitting the channels? Does splitting the channels eliminate the 'extra' green component of the bayer matrix? And what is promoting the red color? The object's coloration? My home site is bortle 2/3, so little light pollution, though some of the images were taken with a lot of moon, if that makes a difference.
 
Just a guess, but...
Integrating each channel separately will generate different normalisation for each channel, potentially changing the balance. This is probably then compounded by the effects of STF on the different balanced images.
 
Thanks again Fred! Because I have to see things for myself, I now have three integrations (regular OSC, split then combined linear fitted channels, and same but without LF) plus ABE and DBE variations to process in parallel.....ha!
 
You must log in or register to reply here.
Back
Top