New Feature : Stellar Colour Calibration for Live Stacking

[MarMax]

Robin,

I was reading Roel's post about this tool and the extreme blue channel after running it. I'm also finding that with "auto stretch" checked the histogram is very aggressively stretched and typically blue heavy. You've indicated that this is how the R-G-G is calculated based on the star colors. And also that running auto stretch and/or auto color balance will defeat the purpose of the tool.

What I do to determine the color balance for the background is zoom to 200% in an area without nebulosity and adjust the color sliders to achieve an R-G-B balance. Doing this after running SCC will typically require a significant lowering of the blue channel. Along with these adjustments also comes changes in the black level and mid level settings.

At the heart of my histogram tweaking is usually a black point setting very close to the average histogram peak. And as I adjust the R-G-B balance downward to reduce the color noise this moves the histogram left so the black point moves right of the peak. This brings up some questions on how to reduce the number of cats to be hearded in this process. We have RGB color balance, RGB background subtraction and black, mid and white point settings to deal with.

Regardless of the different types of adjustments (for me anyway) the histogram black point is at or just slightly left of the histogram peak (average - white line). Would a check box that sets the black point at the average histogram peak be something that is doable or have any value? I'm typically using histogram stretch values of 3 and 4 so this may not work for lower levels of histogram stretch (where users would move the black point more left of the peak).

Mike

[admin]

Hi Mike,

let's see...

If you do the colour calibration, then the values that are set with the most confidence afterwards are the R/G/B colour balance sliders to the right of the histogram - those values come directly from the calibration calculations.

On the other end of the scale, the stretch that is applied with the calibration is pretty arbitrary - at the moment I have it set up to be quite strong, as I always find that the normal auto-stretch isn't strong enough for most targets. Feel free to change the stretch after calibration in any way you choose - it doesn't undo any of the good work of the calibration. You can turn off the auto stretch option in the calibration tab, although that generally means that you need to adjust the stretch manually after each calibrate as the black point will move around, etc.

In the middle in terms of confidence/accuracy are the values used to set the background subtraction - SharpCap calculates them from the histogram for each channel, but the calculation is nowhere near as sophisticated as the calibration calculations for the colour balance. I'm fairly sure that most of the cases where we are getting odd colour casts after calibration are due to these background subtraction amounts being out for one reason or another. Working out the reasons is what I have to do next...

So, in your case, I would suggest using the background subtraction in live stacking to get the neutral background colour, rather than the vertical colour balance sliders in the histogrma tab - I think that should leave you with an image that looks 'right' and is closer to the calibrated balance than doing it the other way. One of the problems though is that adjusting the background subtraction to get that balance is a bit tricky right now because it's not simply a matter of aligning the three 'adjusted' peaks in the background subtraction tab - I am going to try to improve this before next week's update, even if I can't get the auto background subtraction calculation sorted.

cheers,

Robin

[admin]

Hi folks,

a new update today - 4.1.13584 - brings two new improvements to Stellar Colour Calibration

1) The ability to use colour data from a higher resolution plate solving index file.

To take advantage of this, install the new version, then go to the SharpCap Plate solving settings. In the settings, check that the high res index is showing as 'Downloaded' *AND* is showing as 199Mb in size. If it is not downloaded, or it is only about 160Mb, use the 'Download' button to get the latest version onto your PC.

Once you have done this, SharpCap should automatically switch to using the high resolution index and the star colour data included in it for the colour calibration (regardless of your plate solving settings which may select a different index). The high resolution index gives at least 4x as many stars with colour data and should help if you have been hitting the problem where the calibration fails due to not enough matched stars.

2) Improvements to the calculation of the colour balance of the background, which should fix many of the issues with a blue or red colour cast to the background that were being seen with the previous version. This introduces a new adjustment to allow some customization of how the background colour is neutralized

Screenshot 2025-08-04 220705.jpg (10.64 KiB) Viewed 4482 times

This adjustment controls the level at which SharpCap tries to balance the three colour channels. When set to the default of 50%, SharpCap will try to adjust the background so that the median values in each colour channels have the same brightness, typically aligning the peaks in the histogram with each other. This seems to work well in most images - it will be particularly suitable in images where most of the image area is dark background (ie the nebulosity/galaxy in view covers a fairly small part of the image).

If the frame nebulosity/galaxy in view covers a large fraction of the image, then turning this value down may help if the default value gives problems with a colour cast to the background. Lower values will tend to align the left sides of the histogram peaks in the three channels together, which may put the peaks themselves at different positions if one peak is wider than the others.

It's worth noting that the old way (before this update) was roughly equiavalent to using a low value (1-2%) for this adjustment, so anyone wanting to go back to the old way can turn this adjustment down.


cheers,

Robin

[admin]

Hi,

just a little further update - there is now a sequencer step to apply the colour calibration to a live stack currently in progress, allowing this tool to be used as part of an automated imaging session.

cheers,

Robin

[roelb]

Hello Robin.

Thanks a lot for the latest refinements in the SCC feature.
In the mean time I had the opportunity to do 3 EAA sessions.
Sadly with the moon as the "disruptor"
I find that the blue color is still prominent.
Also I find that the upper part of the image is bluer then the lower part.
Is this because of the moon skyglow?
I always left the "Dark align at 50 %ile.

Here are the links to my sessions:

https://mega.nz/folder/wDVBXTxB#PdeLYC2MmF7uDpyoWAzh6w
https://mega.nz/folder/QWEBUJiR#-oeWRYufNCbMc0oPy1n_tw
https://mega.nz/folder/kTcQlaAB#semMafoNdNE8twGZq8-RuQ

Thanks in advance for your opinion on these results.

[admin]

Hi Roel,

thanks for sharing the data - I have had time so far to look at some of the stacks and try to figure out what is going on. The most obvious one so far is Sh2-115, but others show a similar pattern.

The first thing to note is that the calibration wants to apply a large boost to the Blue colour balance as the blue seems to be too dim in the image. This calculation seems to be robust, although you could argue about whether the factor applied should be 1.8, 1.9, 2.0, etc - it's somewhere in that range. Once that blue boost is applied and then the background is subtracted out, you see that the blue curve peak in the histogram is significantly wider than the red/green peaks. This is, I suppose, indicative of the extra noise coming from an initial low blue signal level and then fixing that by multiplying up the pixel values.

With the wider peak, it becomes tricky to get a good background colour across the whole image - aligning the peaks still leaves areas where the background is brighter, and in those areas the extra width of the blue peak is tending to give a blue cast to the image. If you do more blue subtraction at that point (manually) then instead you get more areas that are slightly darker than average in the background getting a yellow/red tint (due to lack of blue). I don't suppose that the bright background along with the vignetting of the corners is helping. In particular, although we can subtract out the average value of the bright background, the *noise* on that large average value remains and cannot be removed - that contributes to wider than normal background peaks when the light pollution is bad, making the whole situation more noticeable.

One possible approach would be to choose a bluer option for the reference white - for instance an F0V star rather than 'default' - that reduces the blue boost amount, which then reduces the amount of boosting of the blue noise and will generally help. Unfortunately when I tried this with the Sh2-115 image, I bumped into an issue. That sort of reference white gives a crosshair down in the left hand tail of the B-G graph, and the fit to the data in that region isn't brilliant. I think I ought to give more weight to the data points near the 'white reference' value to pull the line closer to the data in that region - I will investigate doing that in a future update.

cheers,

Robin

[roelb]

Thanks again Robin for your extensive explanation.
In the mean time I discovered that I did not used the "Linear Gradient Removal", because I thought that it was not compatible with the new SCC.
I used the "Folder Monitor Camera" to redo stacking with LGR on and that gives overall better results.
I do also believe that the bright background due to the moon has negatively affected the results.
The next days the moon will be no longer an issue, so I will redo some objects to compare with the earlier ones.
Yesterday's session was already better: https://mega.nz/folder/QD0hhbzB#w1gA583tB21qikXlSpGNmQ.
I'll keep you posted with newer results.
Thanks again for your efforts.

[admin]

Hi Roel,

glad that the fainter moonlight is helping - are you using any filtering for these images? Just wondering if one of these dual/tri/quad band filters might help for some of the targets (obviously mainly the nebulae). I had a go myself with a quad band filter a few days back and did the same target both with and without the quad band filter - the SCC gave similar final results in both cases, which I think is promising.

I had a look through the new images you shared, which do look better, but one of the Barnard148 images struck me as odd - a clip from it below

Screenshot 2025-08-14 110618_c.png (524.58 KiB) Viewed 3802 times

There are various dotted lines in primary colours here - you'd think they would be hot pixel trails, but that doesn't make sense as they are not all parallel to each other (even with field rotation, there would be a turning pattern in the trails, not the sort of random direction effect seen here).

I can't quite work out what is going on - you might think they would be cosmic ray trails or similar, but then why would they be dotted, and in one colour rather than affecting all colours. Weird...

cheers,

Robin

[roelb]

Robin, just to make sure:
I never use any filters with my ZWO camera's.

[roelb]

I'll take a look in my recordings for those weird trails.
Could it be connected with the use of the amount of hot pixel removal?