"True" Color?

[dmcclain]

I just ran into this website from an imaging scientist working for NASA and ESA. http://www.clarkvision.com/articles/color-of-stars/

What he says makes a lot of sense, and when using my Canon 6D I get near to his colors, but not quite. But on the ATIK 490 CFA I get huge amounts of hard red, which drives in the opposite direction from his recommendations.

From what I can see, PI is *not* doing histogram equalization, as he criticizes, to obtain its STF stretching. It appears that the scenario recommended does the following:

1. Background uniformization - attempt to align the lower tail of the background distributions in the color channels by separate subtraction of small constant amounts.

2. Color Calibration - perform a statistical estimate on stars in the image and individually scale the color channels after background uniformization to make the statistical average star color appear white.

3. STF - Perform a nonlinear stretch of the adjusted scaled color channels so that the background histogram as a mean of 0.25 and approximately 3 sigma = 0.25.

It is that last part (3) that might be causing the departure from real (?) colors. I can certainly agree with the actions in steps 1 and 2, preserving linearity in the base image. How do you know that making that background histogram in the manner indicated is correct?

I'm confused, clearly. Any advice is appreciated.

[dmcclain]

On closer reading of Clark's color of stars data, using derived maps from the Tycho catalog, he indicates that the true average (local) star is generally more yellow to orange - not white.

I also read all the stuff I could on PI, suggesting that the cores of galaxies should be white on average. Hard to argue against that too.

So I wonder if we are trying to misuse the color calibration on local star fields? Perhaps PI was aimed primarily to support galactic research, and its notions of color calibration were aimed in that direction.

Is it possible to perform a color calibration more appropriate to local arms of our own galaxy where stars are generally less white?

[dmcclain]

AHA! I just found out about AssistedColorCalibration and Zbynek's very helpful tutorial! Don't quite understand the math yet, but this definitely helps a lot. Tuned up on the core of M31 and now M42 looks closer to correct.

After going through the exercise on several images, I now understand what is going on. Like a monkey, I had learned one trick and applied it everywhere -- using BackgroundSubtract followed by ColorCalibration on every image. That is not correct.

I now know the proper way is to calibrate the color channels on a galaxy shot, then use those coefficients in every other image first, followed by background subtract, and constrained stretch. In fact, if you have a decent galaxy image, you can let ColorCalibration preceded by background subtraction to estimate those coefficients very closely, even though the order of application of the tools is backwards. The background is very slight, and the effect is minimal. I found that you could iterate if necessary (probably isn't). My iteration in the second round showed a change of 0.001 in only the green channel, so the first estimate from ColorCalibration was probably good enough.

Everyone needs a good galaxy image to calibrate. Use ColorCalibration on that in auto mode, and thereafter use it only in manual mode with the coefficients obtained from the galaxy image.  And use that ahead of background subtraction.

Boy what a difference that makes !!  (I must say the process of learning PI comes in dribbles due to the scattered nature of the documentation. But once you search out a problem, this tool really rocks!)

[pfile]

i think if you want to "go the extra mile" as it were, you probably want a reference image that was captured under similar circumstances (elevation, atmospheric conditions) in the same part of the sky as your target - obviously at a different time of the year. i'd think that you could use a starfield, as long as there are enough stars such that the average is white (that is, not stars that are obscured by dust, etc.)... unless you can find a big enough galaxy somewhere along the path of your target.

rob

[dmcclain]

Roger on that. I have been successfully using the newfound understanding against all my prior images. It works extremely well.

I even did a consistency check - I calibrated against a long (for me anyway) 3 hour stacked image of M31. Then I found a 1 hour stack on M33 on which I used a manual color balance derived from the scalings shown for the M31 image in auto calibration mode. Sure enough, M33's core came out nice and white. For me, that was a sanity check on my own understanding of the method.

Up to this point, I had been mistakenly using auto color calibration on all images... with the expected result that one throws up their hands and states that color in space is just a figment of our imagination anyway.

[Rainer]

Hi,

Great article 

Now after reading that it came to my mind when doing narrow field ( 1' minute or less ?) imaging in colour RGB what is the real background colour of the deep space ?

RGB values of 0,0,0 or perhaps RGB 5,5,5 or even perhaos higher like RGB 20,20,20 ?

My thinking comes from the fact that if the universe is 13'000'000.000 light years wide or deep or whatever, there must be a lot of stars twinkling light, but we are not able to detect it and therefore the deep space colour can not be RGB = 0,0,0

¿ or ?

regards Rainer