White balance advice

[Jeff2011]

I am having a lot less issues with color balancing as I used to, but I would like to get a better understanding of how StarTools does it and get advice on how I can improve my process.

My current method is to use a custom white balance on my modded DSLR and then have DSS use the cameras white balance. Then in StarTools in the color module I usually zero any changes that StarTools makes to the color channels and increase the saturation and adjust the top and low end sliders until I get something I like. If it still looks wrong I then may tweak the red, green, or blue to get something that looks right.

When I first click on the color module with a full mask, StarTools does a white balance on the whole image and adjusts the color sliders. How is it doing this and should I keep these results. One thought I had was to locate a known white star that is not over saturated and click on it to do the white balance. Will this technique work and are there any others that I can use.

Thanks,

Jeff

[admin]

My current method is to use a custom white balance on my modded DSLR and then have DSS use the cameras white balance. Then in StarTools in the color module I usually zero any changes that StarTools makes to the color channels and increase the saturation and adjust the top and low end sliders until I get something I like. If it still looks wrong I then may tweak the red, green, or blue to get something that looks right.

In StarTools you will get best results if you don't pre-color balance your data at all. The reason for this, is that color balancing meddles with the noise levels per-channel and hence also the luminance derivative.
To keep noise levels at their lowest level possible, let StarTools color balance your data. Because StarTools separates color processing from luminance processing, it is able to modify color balance after it has separately extracted the unmodified/virgin luminance data, allowing for the best of both worlds.

Additionally, StarTools can give a small boost to your signal if it may further assume that the data originated from a OSC or DSLR with bayer matrix.

When I first click on the color module with a full mask, StarTools does a white balance on the whole image and adjusts the color sliders. How is it doing this and should I keep these results.

StarTools looks at the average color of all pixels that are set (green) in the mask, based on the assumption that the universe is on average white. This is especially true in wide fields.

One thought I had was to locate a known white star that is not over saturated and click on it to do the white balance. Will this technique work and are there any others that I can use.

This technique is indeed one you could use (use a G2V star as a white reference).
You could also create a mask with just all stars selected (the average star population comes out white), use a suitable nearby galaxy as a white reference (be sure to select both core and outer ring), or use a G2V star as a white reference. Put the object(s) in a mask, click 'Sample' in the Color module. This will set the ratios correctly. At this stage, the mask will cause the image to be processed selectively, so in the mask editor, Clear & Invert the mask to select the whole image again and return to the Color module. Now the whole image will be color balanced.

[Jeff2011]

Thanks for the detailed explanation Ivo. Using the camera white balance was an easy way for me to get good looking colors but was costing me noise. I will give your suggestions a try.

Taking this a step further, would it be possible to modify StarTools to solve the stars and locate G2V stars automatically. Madratter pointed me to a program called Regim that did this, but I think the solving was done through a website. Perhaps that would be a licensing issue since your application is for sale and Regim is free. Just a thought.

edit: After rereading the Regim document, it solves for up to 200 stars of known color indexes and uses those stars to balance the colors. For a small FOV image there is a good chance no G2 stars will exist.

[admin]

Taking this a step further, would it be possible to modify StarTools to solve the stars and locate G2V stars automatically

Definitely an idea for a future version.

As for using Regim, use Regim on the Wiped, linear (unstretched) data only. Use the new Restore function if you're not sure how to get at the data in that state. Make a note of the color balance factors/multipliers that Regim comes up with and use those in the Color module.
Do not use Regim to color balance the linear data (for noise reduction reasons) and do not use Regim to color balance the unwiped data either (it will give wrong results).

[Jeff2011]

Thanks for the tip Ivo. I had been wondering how I could incorporate Regim's color balance without affecting the StarTools noise reduction logic. Your explanation makes sense to me. I will give that a shot and let you know how it goes. It would be an awesome enhancement to add this type of color balance logic into StarTools. Thanks again for the quick and well thought through response.

[Jeff2011]

Here are my results. I just cropped out a small part of my image so I could show at full resolution.

color balanced before StarTools:

Processed by applying custom white balance before StarTools

CROP_OLD.jpg (332.69 KiB) Viewed 11774 times

color balance in StarTools:

Processed with color balance in StarTools

CROP_NEW.jpg (289.66 KiB) Viewed 11774 times

The only thing I was not sure of was how to apply the Regim color balance values. After I did the wipe and cleared the stretch, I saved the file and loaded it in Regim. It successfully solved and reported a color correction of:

05.01.2014 19:59:15 : (LOG) fr=0.7176365022058852
05.01.2014 19:59:15 : (LOG) fg=0.9318704502427896
05.01.2014 19:59:15 : (LOG) fb=1.0

Do I enter these numbers in the Develop Module or the Color module? If in the Color module, how would I do that? In my example I entered these into the Develop Module as 72% red, and 93% green and did not even go into the Color module.

Thanks,
Jeff

[admin]

Hi Jeff,

The only thing I was not sure of was how to apply the Regim color balance values. After I did the wipe and cleared the stretch, I saved the file and loaded it in Regim. It successfully solved and reported a color correction of:

05.01.2014 19:59:15 : (LOG) fr=0.7176365022058852
05.01.2014 19:59:15 : (LOG) fg=0.9318704502427896
05.01.2014 19:59:15 : (LOG) fb=1.0

Do I enter these numbers in the Develop Module or the Color module? If in the Color module, how would I do that? In my example I entered these into the Develop Module as 72% red, and 93% green and did not even go into the Color module.

Don't use the luminance mixer in the Develop Module (it merely changes the channel contribution to the luminance component - only useful in a handful of planetary imaging cases).
A little bit of conversion is needed to come up with the percentages that StarTools requires.

Simply take the lowest of the three numbers (in this case fr=0.7176365022058852), divide all three numbers by this number, and multiply all three results by 100%.
ex. 0.7176365022058852 / 0.7176365022058852 = 1.0 * 100% = 100% red
ex. 0.9318704502427896 / 0.7176365022058852 = 1.298 * 100% = 130% green
ex. 1.0 / 0.7176365022058852 = 1.393 * 100% = 139% blue

[Jeff2011]

Thanks Ivo. I figured the Develop module was the wrong place and I have been trying to work out a translation of the Regim numbers. With regard to your scale in the color module, the value of the RGB sliders go from 1 to 10. Is 1 equivalent to 100%? So with your calculated numbers I would leave red at 1 and then set green to 1.3 and blue to 1.39 assume I had the sliders set to increase?

A few other questions concerning DSS settings. Is it OK to align the RGB channels in DSS or will this mess with the StarTools noise reduction also? I am also using kappa-sigma stacking for light frames which from my understanding requires background calibration. I am currently using per channel background calibration. Perhaps it would be useful if you had a brief tutorial on recommended calibration/stacking software settings compatible with StarTools.

Thanks,

Jeff

[admin]

Thanks Ivo. I figured the Develop module was the wrong place and I have been trying to work out a translation of the Regim numbers. With regard to your scale in the color module, the value of the RGB sliders go from 1 to 10. Is 1 equivalent to 100%? So with your calculated numbers I would leave red at 1 and then set green to 1.3 and blue to 1.39 assume I had the sliders set to increase?

Yes, sorry, my bad! Disregard the whole percentages thing!

A few other questions concerning DSS settings. Is it OK to align the RGB channels in DSS or will this mess with the StarTools noise reduction also?

RGB align to counter star color shift is totally fine. (it's preferred actually!)

I am also using kappa-sigma stacking for light frames which from my understanding requires background calibration. I am currently using per channel background calibration. Perhaps it would be useful if you had a brief tutorial on recommended calibration/stacking software settings compatible with StarTools.

Thanks,

Jeff

Use "Per Channel Background Calibration" (which should not upset the color balance).
I did start on a guide for DSS but got bogged down by a number of DSS bugs that made operation inconsistent (or sometimes downright unusable) between versions. I decided to wait until the current DSS beta version is declared stable.

[Jeff2011]

Thanks for your quick reply. I am glad to hear that I at least I got the other DSS setting right.

However, I am running into a few things.

I am using the latest beta version of StarTools and I seem to have an issue using the new Restore feature. If I restore to Linear and Wiped, I get what looks like my original unwiped image. I know this because I test it by going into development and do a stretch on it and I still see the gradient LP. If I restore to Linear - Wipe - Decon, then I get what looks like my wiped image unstretched even though I did not do a decon. If I now save this, it appears to save my original unstretched unwiped image instead. The only way I have been able to save a wiped unstretched version is if I go back to development after a wipe and select global restretch and then don't change the sliders, keep and save. That appears to save a linear wiped version.

The results I am getting from Regim do not appear to be repeatable. I take my same linear wiped image and run it several times closing all of the windows in between each run. Many times it fails to match stars but when it does there is a good bit of variance between each run. I also tested the Regim results against masking the stars and doing a white balance in StarTools as you suggested earlier and I am really not seeing much difference between that method and using Regim.

I still don't think I have the color balancing quite right yet, but I am working on it.

Thanks,

Jeff