Superpixel = supergreen

[TinySpeck]

I'm fooling around with some old data, taken with a Canon DSLR.  I don't have bias, flat, or dark frames for it.  I stacked without image calibration, and used SuperPixel debayering for the first time.  My result looks reasonable, but it's way too green.  I've tried BackgroundNeutralization followed by ColorCalibration using the image stars as a white reference, and also PhotometricColorCalibration.  Both results are nearly identical, with about 2x the green there should be (i.e. the green hump in the image histogram is about twice as high as the red and blue).

Is there something fishy about the Superpixel method?  I've read here in the forum that it does divide G intensity by two, which sounds right, but is there something else I need to do?

Thanks for any help!

[pfile]

when you do the BN step, what does the console show for the multipliers?

rob

[TinySpeck]

Hi Rob,

BackgroundNeutralization: Processing view: Image_a
Evaluating background: done
* Background reference:
B_R : 1.60213e-01
B_G : 8.32530e-02
B_B : 1.25977e-01
Applying background neutralization: done

So G is about half of the R and B.

[TinySpeck]

Here is the image after BN and CC.  It looks just the same after PCC.

Normally BN lines the RGB humps up along the X axis and CC brings the peaks to roughly the same, but not this time.

[pfile]

ok, i see - i thought you still had an overall green cast after the BN. but it doesnt look like it, instead you just have green halos.

i don't really have a good explanation for what's going on - does your white reference mask seem to encompass the entirety of the stars when you run ColorCalibration?

have you tried SCNR to green to see what you get?

rob

[TinySpeck]

It's not just green halos.  The background cast is green too.

Yes, my background and white ranges for CC are correct, and I've experimented with them too.

I can reduce the effect a little with SCNR applied to green, but CC or PCC shouldn't require this.

What bothers me is that even PCC gives this same result.  The only color-related parameter you can adjust there is your background neutralization range, so there's really nothing you can mess up.  The color should just be correct.

[pfile]

can you post the integrated image before any CC / PCC?

rob

[TinySpeck]

Here is a link to the .xisf file which resulted from stacking:
https://drive.google.com/file/d/1eNR3FS084KxjMEYwMakg7hYcpUMD_icq/view?usp=sharing

That's not permanent, but I'll leave it up there for a long while.

[pfile]

ok i looked at this a little bit... the image pre-DBE looks pretty good, smooth background and mostly grey. post-DBE there is a green cast. i think something like this is pretty common, as DBE is removing the sky signal in the process of removing the vignetting, and so you're left with just the sky noise sitting there in the background. so i'm not sure if this is specifically about superpixel, but maybe the interpolations going on with regular debayering would reduce the mottling somewhat.

do you have flats for this data?

anyway after doing the DBE and following up with SCNR to green at about 70% i was able to get something that looked reasonable. unfortunately the "pink star cores" problem is there since this is 14-bit data. i messed around with the Repair HSV separation script a little to fix the star cores but in the end i had to merge that data in using a mask and pixelmath since the script really flattened out the star cores completely.

rob

[TinySpeck]

Hi Rob,

Thanks for looking at this.  How did you get pre-DBE?  What I posted was just an image, without the history.

I didn't actually do DBE, either.  I removed the stars from a clone and smoothed the result, then used that as an artificial flat and divided it out of the original image.  I don't have a flat for the image. 

I can see how the artificial flat might result in a color bias.  But that shouldn't matter, should it?  CC or PCC should start from whatever  color-biased image you have and pull the colors to where they need to be, unless I'm misunderstanding how they work.

I'm reluctant to use SCNR or Curves to fiddle with the color until I like it; I'd rather use a white reference of some sort (CC or PCC) and see what the real star and background color is.

I'll experiment with a different or neutralized artificial flat this evening.  But I still don't get why CC and PCC ended up with such an obviously uncorrected result.

[pfile]

well, what i meant is that the background did not look green to me until i did DBE, which is what i always do when there is uncorrected vignetting. i had assumed you had done the same since the background looked pretty good without the DBE. i dont know why we are seeing different things.

you are right that the color bias from any flat should not matter, color calibration should fix it.

at any rate personally i dont feel that SCNR is such a radical approach - it is just another way to fix a particular type of noise.

rob

[TinySpeck]

OK, I get it.  That's right, the version you got was pre-flattening (DBE).

Something that generally fools me about PI is that the background can look worse after flattening (gradient equalization), if you use automatic STF before and after.  With a gradient in your background (pre-flattening) there is a larger dynamic range to the image and the background consumes more of it, so it looks smoother.  After flattening there is less dynamic range and the background is all about the same level so small variations look worse.  When I've really pushed the linear background noise reduction it gets worse and worse looking!  I have to remind myself that the automatic STF does that.

[TinySpeck]

Well, this is embarrassing...  My green bias goes away when I link RGB channels in STF.  Normally I get better results (more balanced-looking color & background) when I unlink them, but not this time.  I guess I need to learn what's going on behind the link/unlink better.

When I stretch this image I actually get a slightly reddish background result, not green, showing that there isn't a green bias in the linear image after all.

[pfile]

ah... ok. right, after any kind of color calibration you need to use STF in linked channel mode, or else STF "undoes" the calibration. STF just looks at the statistics of the image to compute the display histogram and if it is in unlinked mode it computes each channel independently of the other. this can be a good thing to preview what un-color calibrated data would most likely look like if it were calibrated, but once the casts have been removed you don't want to do that.

rob

[TinySpeck]

Thanks for this explanation and the help with my problem, Rob.  This improves my understanding of the linked/unlinked STF.