Unusual Results from Photometric Colour Calibration

[dhb2206]

The image: https://www.dropbox.com/s/nxtxzgrntx8afj8/181108%20Drizzle.zip?dl=0

I've come across some rather strange happenings with one image in particular, it is certainly nothing special, but thought this merited a post. I'm using just a DSLR and camera lenses at present and shortcutting all over the place as well as experimenting. So this time I was seeing whether my skytracker would cope with two minute exposures, the answer was "sort of" - 50% of the exposure went in the bin. Anyway, I digress. I followed my usual workflow; debayer, register, integrate, drizzle integrate, DBE, BN and I come to PCC. I ran PCC without a solved image and wound up with a very red result with white balance and background readings of:
* Color transformation functions:
R-G = +9.740790e-01 + 8.713840e-01*(Sr-JV) ± 4.639148e-02
B-G = +2.756727e-01 + 4.287585e-01*(JB-JV) ± 4.129395e-02
* White balance factors:
W_R : 1.0000e+00
W_G : 4.8952e-01
W_B : 8.3530e-01
Applying white balance: done
Evaluating background reference: done
* Background reference:
B_R : 9.16751e-05
B_G : 5.58175e-02
B_B : 9.53854e-02
Applying background neutralization: done
02:53.94

So, red is obviously very wrong and I undid PCC.

Run #2 I plate solved first for M45, no problem (Focal 171.53 and Pixels at 1.9). Then ran PCC again but received a very blue image this time (note that I suppressed the saturation threshold to 0.8 in this run) and received:
* Color transformation functions:
R-G = +3.833681e-01 - 9.225998e-01*(Sr-JV) ± 5.962888e-02
B-G = +1.069563e+00 - 3.878159e-01*(JB-JV) ± 4.107895e-02
* White balance factors:
W_R : 8.3131e-01
W_G : 4.8125e-01
W_B : 1.0000e+00
Applying white balance: done
Evaluating background reference: done
* Background reference:
B_R : 9.50452e-02
B_G : 5.47973e-02
B_B : 9.64830e-04
Applying background neutralization: done
02:27.95

This time blue is very obviously wrong.

How do I fix this one?

Cheers,

David

[sharkmelley]

I don't understand your statement that "red is obviously very wrong".

I downloaded your file (which is massive by the way) and more or less duplicated your first run.  Here are my results:

* Color transformation functions:
R-G = +9.740677e-01 + 8.713659e-01*(Sr-JV) ± 4.845947e-02
B-G = +2.728715e-01 + 4.308724e-01*(JB-JV) ± 4.088340e-02
* White balance factors:
W_R : 1.0000e+00
W_G : 4.8952e-01
W_B : 8.3431e-01
Applying white balance: done
Evaluating background reference: done
* Background reference:
B_R : 9.16751e-05
B_G : 5.58179e-02
B_B : 9.52719e-02
Applying background neutralization: done
02:57.58

I think your problem is that the background level is so high that the background neutralisation step fails with the default parameters but that's only a guess.  I see the same problem as you - the white balance has been applied but the background neutralisation has not.

So after running PCC try running BN manually and set the upper limit to 0.25 instead of the default of 0.1.
Now run HistogramTransformation setting highlights to 0.4 to deal with the pink in the core of the bright stars.
ArcsinhStretch (in the real time preview I did "Estimate black point" and I used a stretch factor of 140) then gives the attached result.  ArcsinhStretch was used in this instance because it faithfully reproduces the colours during the stretching.
The star colours and nebulosity colours look pretty much OK to me.

Mark

[dhb2206]

Thanks Mark.

I don't understand your statement that "red is obviously very wrong".

The fact that it was singular was all, red in one instance, blue in the other. Another penny drop time - is one of RGB always 1? After all you do need a start point.

I downloaded your file (which is massive by the way) and more or less duplicated your first run.  Here are my results:

Yes, the files are massive so I'm wondering if I'm missing a trick there - each frame is 20MB, post debayer these are 10 times that size at 200MB each and then the drizzle doubles that.

As it happens I avoided colour calibration altogether in the end and just scnr (as the greens always dominate) gave a pretty natural colour anyway. Thanks for the arcsinhstretch tip, I'm still very much at an early stage in the learning curve.

David

[dhb2206]

Aha, the penny drops. Not so much a bug, more a mug that didn't twig!

So thanks to Mark I tried again, plate solved, then BN at 0.2 (I hadn't noticed my background was rumbling away at 0.1+) then PCC with pcc bn set to 0.2 as well. I still had singularity on Blue, but the output image was far more natural this time.

I should do a bit more reading up, but even with a degree in maths, some of the techy stuff goes right over my head!

David

[sharkmelley]

Of course! The BN upper limit can be set within the PCC.  I hadn't spotted that!

Also it is always the case that one of the channels (the weakest one) will also be assigned a white balance multiplier of  exactly 1.0
This prevents the white balance operation clipping any bright highlights and therefore gives full control to the user.

Mark

[dhb2206]

Also it is always the case that one of the channels (the weakest one) will also be assigned a white balance multiplier of  exactly 1.0

That penny dropped as well, I was just modifying my script above when your reply came in! I'd never really paid much attention to the stats before, but then you don't when things "work", it is only when the spanner appears in the works that you start taking a real interest.