[argyllcms] Re: Scaled Rec709 TRC question

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Tue, 15 Feb 2011 10:14:21 +1100

János, Tóth F. wrote:
But it sounds much easier to set up those three ambient view condition
presets with fixed curve parameters.

You could probably approximate what CIECAM02 does using a standard
power-curve-with-straight-segment, but to do that you'd have to sample
the CIECAM02 curve and then use a fitting algorithm to derive the
parameters. Is that somehow easier ?

Of course I just used the whole CIECAM02 algorithm because I had
it available (and I think it is in lcms too), but you could
tease out the luminance processing and figure out a simplified
equation, but it won't be a power-curve-with-straight-segment.

Otherwise, I can't really believe in CIECAM02 here because it's much younger
than Rec709. (There is no way that Rec709 has anything to do with CIECAM02.
At least officially...)

They are completely different things. Rec709 is just a standardisation of
an video encoding. CIECAM02 is a human color appearance model, so it's
specifically designed to model color appearance, rather than it happening
as a side effect, which is what has happened in TV standards.

So, I think: If the "scaled inverse encode curve" is really meant to be used
on the end-user side, then there should be a commonly accepted/suggested way
for this scaling method. (Which had to be exist long before CIECAM02...)

No there is not. The color appearance transform effects of the mismatch
between the TV encoding and actual display response hasn't been widely
understood. Presumably the original TV engineers tweaked the encoding
curve to give the right looking response on a CRT display, and this
winds up corresponding to an encoding gamma of about 2.2. A typical CRT
has a gamma of 2.4 or so, giving the studio bright to living room dim appearance
adjustment. Read this 
and this http://www.poynton.com/notes/color/GammaFQA.html,
or better yet, get a copy of Charles Poyntons book:

And another concern of mine that this curve is well defined but otherwise
"hard to scale". It wouldn't make me think that I should use it in a scaled
form. (Or where are the scaling formulas and/or the suggested curve
parameters then...?")

I'm not quite sure what you mean.

Graeme Gill.

Other related posts: