Re: [Lcms-user] Creating equivalent matrix-shaper and 3D LUT profiles, problem with BPC

[Graeme G. <gr...@ar...>]

mar...@li... wrote:

Hello Marti,

> Sure, 16 bits is a plenty of room, but if you use 16 bits to encode L* and a
> big part of the domain are highlights over 100, then you could end 0..100 to
> be encoded using 9 bits and this is still not enough for a decent accuracy. 

An available HDR display is lucky to hit 2000 cd/m^2. If you pick a diffuse
white of (say) 200 cd/^m, and the max is encoded as L* = 100, then the
diffuse white ends up at L* = 37.8. With 16 bits that leaves 24773
perceptually encoded levels SDR range or 14.6 bits. That's a resolution
of 0.004 dE, more than enough.

Worst case XYZ PCS is not so favourable of course - a 10000 cd/m^2
display running at a diffuse white point of 100 cd/m^s would only
have a little over 9 bits of linear encoding available. This would
have visible banding in black. So yes, either a ICC V2 16 bit L*a*b* PCS
or ICC V4 MPE XYZ PCS floating point to encode source standards.

> Another issue is what to do with those highlights. A tag with HDR diffuse
> white would help to do a coarse clipping, but if you want some gamut mapping
> to bring highlights into SDR gamut, you have to redo everything.

I really don't imagine doing this in the profile itself. Too inflexible.
Better off doing this in the linking or execution of the source to destination
transform using non-ICC machinery.

> Then, since I could not discard such big number of profiles, I tried to
> imagine a way to automatically do the rescaling from different PCS. And
> found BPC to work fine for that. So, the idea behind this change is to
> increase inter-operability and fix broken profiles.

It seems to me that BPC makes black worse almost as often as it
makes it better though!

Case in point with lcms BPC:

Using this RGB printer profile: <https://www.argyllcms.com/bpctest.icm>

transicc.exe -t1 -isRGB.icm -obpctest.icm
Enter values, 'q' to quit
R? 0 0 0
R=25.8949 G=5.3969 B=0.0000

icclu -ff -ir -s255 bpctest.icm
25.894900 5.396900 0.000000 [RGB] -> Lut_A2B1 -> 5.693560 0.252739 -3.343661 [Lab]

transicc.exe -b -t1 -isRGB.icm -obpctest.icm
Enter values, 'q' to quit
R? 0 0 0
R=41.5914 G=9.5681 B=0.0000

icclu -ff -ir -s255 bpctest.icm
41.591400 9.568100 0.000000 [RGB] -> Lut_A2B1 -> 8.089132 -0.308324 -2.011786 [Lab]


At least the lcms BPC doesn't seem to wreck the perceptual black on this one:
(unlike Adobe's always on BPC in Lightroom)

transicc.exe -b -t0 -isRGB.icm -obpctest.icm
Enter values, 'q' to quit
R? 0 0 0
R=19.3152 G=11.3658 B=24.4436

icclu -ff -ir -s255 bpctest.icm
19.315200 11.365800 24.443600 [RGB] -> Lut_A2B1 -> 4.296325 -0.431107 -5.422716 [Lab]

Cheers,
	Graeme.