Our rec2100-hlg is wrong

[svgeesus]

I see from the convert app that the results for rec2100-hlg are very wrong.

I thought I knew how to fix it, looking at BT.2390, but I tried it and I don't.

(Note that BT.2390 expects the linear-light intensity range Edash to be 0 to 1 over the HDR range, while our code puts 0-1 over the SDR range so media white is at 1. Even accounting for that, I still couldn't fix it).

It seems from this email that things are more complex than I thought.

Explorations meanwhile are in a separate repo and I want to get some expert review from BBC folks before that comes back to color.js.

In particular there will probably need to be an options object with

• Lw, the display monitor peak luminance (default 1000)
• Lb, the display monitor black luminance (default 0.05)
• Lamb, the ambient luminance (default 5)

to cope with non-reference monitor and non-reference ambient illumination.

Meanwhile here be dragons and don't use our rec2100-hlg for anything important.

[LeaVerou]

You might want to add a note to the JSON that generates our color space docs.

[svgeesus]

Note added in f1fa0ed

[svgeesus]

Now fixed and gives the results expected from ITU-R BT.2390-10 page 25:

Report ITU-R BT.2408 indicates that, for HLG HDR, diffuse white should be set at a signal level of 75%. This can be configured by making the output from an 18% grey card correspond to a signal level of 38%, rather than the 42.5% stated above.

• white gives color(rec2100-hlg 0.75 0.75 0.75)
• color(rec2020-linear 0.18 0.18 0.18) gives color(rec2100-hlg 0.3782 0.3782 0.3782)

[svgeesus]

The scaling factor assumes an overall system gamma of 1, and should be a default which is adjustable for different viewing conditions. For now, it is correct for standard viewing conditions.