How Do I Scale Luminosity?

[wswartzendruber]

Let's say I've got a HDR10 movie which grades reference white at 100 nits. Well, let's also say I want that level of brightness to be at 203 nits instead.

Right now, I'm naively multiplying linear display brightness by the required factor (203/100). This looks mostly fine, but BT.2390 is talking like I also might want to consider gamma correction here.

Now, this makes sense. The artistic difference between 100 vs. 50 nits is not the same as 10,000 vs. 5,000 nits, which would be the result of my current approach.

So how do I perform this gamma correction? I can't seem to find anything on it.

[wswartzendruber]

I'm noticing that BT.1886 doesn't seem to take this into account when converting BT.709 relative brightness to a given nit value. Ratios between relative brightness levels remain consistent.

If gamma correction indeed needs to be done, then BT.1886 shouldn't be doing this.

[quietvoid]

In this case you're on the PQ scale.
In the [0, 1] scale: 100 nits = 0.508078, 203 nits = 0.580689, etc.

I don't think it would look good with a simple multiplication, though.
You have to take into account colors and how they're affected by the new brightness.

[wswartzendruber]

I'm scaling the display linear RGB channels separately.

[wswartzendruber]

I think what i want is in BT.2446-0 section 4.1.2.

"Gamma compensation and scaling"

[benwaggoner]

In general you don't want to do math on PQ code values. They're even more nonlinear than gamma, and weird things happen.

Linear scaling of nits isn't that bad a way to adapt to brighter ambient light conditions, since PQ itself is defined relative to a 10 nit ambient. It's certainly better than trying to scale or add to PQ code values.

[wswartzendruber]

Is there even a correct way to do this, or is it subjective?

Right now my procedure is:

1. Take display linear RGB.
2. Convert to XYZ.
3. Convert to yxY.
4. Scale Y value.
5. Convert back to XYZ.
6. Convert back to display linear RGB.
7. Gamma compensation of some kind???

I'm also staring at CIE L*a*b* and wondering if that can't help.

[wswartzendruber]

I just did some experimentation with multiplying linear light and comparing that to gamma curves.

Linear multiplication of luminosity does indeed appear to be correct.