Weighting coefficients for luminance curve

From Rawpedia

"Despite showing the R, G and B histogram (merged) in the background of the curve, the curve operates on luminance values, where
Relative Luminance Y = R0.2126729 + G0.7151521 + B*0.0721750 "

But is this formula used for all the working space?

For Rec2020 the correct formula should be:
Relative Luminance Y = R0.2627 + G0.678 + B* 0.0593
https://en.wikipedia.org/wiki/Rec._2020#Digital_representation

for Prophoto I guess:

Relative Luminance Y = R0.288040 + G0.711874 + B* 0.000086

from here
http://www.brucelindbloom.com/index.html?WorkingSpaceInfo.html

Maybe @jdc can shed some light.

I believe it depends on the illuminant used to make the calculations.

You are true, coefficients are different according to the place or processus. But is it important ?

Yes these coefficients come from matrix conversion RGB ==> XYZ, each matrix depend from Colorspace working.
In majority of cases in RT, the most uses is NTSC, after we found sRGB, and sometimes, real Colorspace is take into account

Recall : XYZ has been created in 1936, by works of CIE (Commision Internationale de l’Eclairage). They constated that human perceived more green than red and blue
Some people so called, the green channel, luminance and the blue and red channels, chrominance, but its clearly false, or very very approximative.
We retreived also these appelations in RGB, for demoisaicing.

XYZ are used in white balance, and are the component of “white”, for example D50 X=0.96 Y=1 Z=0.82. X and Z changes with temp, and Y with “tint” (green channel). We see we are far from “luminance”.

Some years after, CIE works on a system, where color will be perceieved nearly our vision: Lab, was born.

In fact, if we want to have a not too bad approximation, it needs according to the place in the process :
a) suppress gamma if need ; b) convert RGB XYZ according to colorspece working ; c) take the cubic racine of “luminance Y” or directly work in Lab mode

But Lab mode is not good enough, some years later, a Munsell space has been create, which correspond to human vision. This space cannot be used directly, but some years ago, I have developed LUTs (195) that convert Lab to Munsell.
You dont’see that but when you checked “Avoid color shift”, these features is enabled fot a* and b* channel

Some years after, is born the real first CAM (Color Apperance Model) - Lab was not really one, and in 2002 Ciecam02 is good.
In RT, Ciecam uses this feature and assures a perception of the “luminance” which here is called either Lightness (J) or Brightness (Q)

You can see that vocabulary is not only semantic, it is a very, very important notion, which leeds to misunderstandings .
What is color, hue, tint - this terms are often synonym…What is luminance, lightness, brightness…etc.

In summary, one can change the coefficients, if the utility is felt, to each developer to examine the code, the change and the possible repercussions

1 Like

@age, if you want the TL;DR, I guess the answer is yes. However, I’m not sure it will make much difference in practice, as @jdc wrote (or did he? :thinking::yum: )

1 Like

@agriggio

Perhaps to verify, but I think low differences :slight_smile:

So even with a change of the coefficients the curves are virtually the same as before.
Thanks for the answer :slight_smile: