How is the luminosity value of the Composite channel calculated?

[Laurence-Baker]

Taking an individual pixel in the Composite channel is it a mathematical mean of the three channels luminosity, or is some other method used?

Thanks

Laurence.

[D Fosse · Accepted solution]

No, it is not the average value. That would result in dark values for yellows and light for purples.

It is very closely related to the L channel in Lab, but it's not identical. Nobody really knows why it differs from Lab L in the first place, because that would have been a very natural reference. No documentation has ever been released. I suspect it's just an ad hoc-modification to make it more visually consistent. After all, it's just for internal use in Photoshop, so it doesn't have to match any standard.

But for most practical purposes you can think of it as Lab L. The main characteristic of this model is that it takes into account the inherent lightness of each color, so that yellow remains a light value and purple remains a dark value.

[Laurence-Baker]

Thank you for your detailed reply. Time to read up on Lab colour space....

[Stephen Marsh]

Fill an image with black, white or a neutral R=G=B value and fade/blend to Color blend mode. RGB based luminosity values are also found on the master curve and layer option blend-if sliders.

The official formula can be found here, for what it is worth:

https://www.adobe.com/content/dam/acom/en/devnet/pdf/pdf_reference_archive/blend_modes.pdf

[Laurence-Baker]

Thank you. I didn't think it would be simple...

[D Fosse]

I always thought that formula was a myth (R=30, G=60, B=10). It would make so much more sense to base it on Lab L. Since Lab is a PCS (profile connection space) in all color management operations, Lab numbers are always available in the background.

So I decided to find out once and for all. I made a composite from the three channels, based on these percentages - and it checks out! The result is indeed identical to Luminosity. Even the histograms are a perfect match (while the Lab L histogram is significantly different).

So I suppose that settles it. Not, in my view, an elegant solution, but as long as it works...

[c.pfaffenbichler]

»I made a composite from the three channels«

Did you use Channel Mixer for that? 

[D Fosse]

Yes, that's one way, I suppose there are others. You need to set the percentages in each channel.

Then compare to a duplicate with a monochrome layer in Color blend mode, which gives you Luminosity. They are identical here (just repeated it).

The funny thing is that I still prefer the Lab L version, as I think it gives a truer representation of the perceived brightness of the original. I still can't understand why they didn't just use that. It still seems to me a so much more elegant and efficient way to do it, instead of this percentage kludge.

But then again, it may be based on the eye's sensitivity to different wavelengths. There's also CIE XYZ, which (in my limited understanding of it) uses something similar to set brightness values.

[c.pfaffenbichler]

As Photoshop predates the ICC-standard I wonder if many of Photoshop’s Blend Modes (especially the »basic« ones) may not also predate what we consider Color Management now. 

[Stephen Marsh]

The 30/60/10 is very close, but not perfect/lossless.

Filling with a neutral and fading to color blend (the inverse of luminosity) does create a truly lossless conversion to RGB luminosity. It is easy to test the differences between the two layer over the original image in luminosity and then using difference blend mode and equalize to compare.

[c.pfaffenbichler]

Doesn’t »Blend_Modes.pdf« indicate 30/59/11? 

[D Fosse]

It does. I just simplified a bit.

Yes, it could well be that blend modes were invented before color management. I don't even know if the earliest versions supported Lab, let alone RGB color spaces. So that could be the explanation. They just worked with what they had, and it sort of worked, so it stuck ever since. Lots of Photoshop properties came about that way.

[scotwllm]

The official formula cannot be found there. Maybe they moved it. If you know where I could find it now, I will be impressed by your generosity, intelligence, and good looks.

[c.pfaffenbichler]

https://printtechnologies.org/wp-content/uploads/2020/03/pdf-reference-1.6-addendum-blend-modes.pdf

[scotwllm]

retracted