Re: [css-color] wider/deeper colors

On Tue, Mar 22, 2016 at 5:34 PM, Florian Rivoal <florian@rivoal.net> wrote:

>
> > On Mar 23, 2016, at 01:56, aaron.www-style-@infinite-source.de wrote:
> >
> > On 22.03.2016 08:45, Florian Rivoal wrote:
> >>
> >>> On Mar 22, 2016, at 11:34, Rik Cabanier <cabanier@gmail.com> wrote:
> >>>
> >>> It could be a new CSS property, for instance: compositing-space:
> "p3"/"sRGB"/"uncalibrated"/url(..)
> >>
> >> Shouldn't that be "compositing space: sRGB | output | CIELab" ?
> >>
> >> * sRGB: legacy "make everything dull" mode
> >>
> >> * CIELab: because if you're compositing things coming form various
> color spaces or wide gamut color spaces, this is how you get the least
> color distortion.
> >>
> >> Say your sources are in a gamut wider than the output gamut, and you
> have semi-transparent very saturated red on top of very saturated blue. If
> you composite in CIELab, then convert to the output gamut, you'll get a not
> very saturated color that may not need clipping to fit the output gamut.
> >>
> >> If you fit the colors to output gamut first, then do composition, the
> result will be different (and worse).
> >>
> >> * output: because if all the things you are compositing do fit in the
> output device's gamut, doing composition directly in that space will be
> close enough to what you'd get using CIELab, but is computationally more
> efficient.
> >
> > Compositing in the output gamut is not really close to compositing in
> > Lab. Try generating a saturated red-green gradient or rainbow in sRGB
> > and in Lab.
>
> Right. I should have qualified that some more. Compositing in Lab gives
> better results, and compositing in the output gamut is sorta of close
> in some, but certainly not all scenarios. Depending on what you're
> doing, it may give be good enough, or maybe not.
>
> If we had no compat constraint nor performance constrain, I'd say we should
> always do composition/color math in CIELab, but we may not have that
> liberty. We should certainly explore whether we have it though, as it would
> be better.
>

No, compositing in CIELab will give you unexpected results.
If you have access to Photoshop try the following:
- create an rgb document (16bit is best)
- create a red rectangle
- put a green rectangle inside, give it 50% opacity
- use the eye dropper on the green rectangle. It measures 50% green and 50%
red, no blue (as expected)
- now switch to the document colorspace to Lab. Do not flatten
- notice how the transparent green rectangle changes color. It now is 78%
red, 67% green