[argyllcms] Re: fwa compensation unexpected results
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 05 May 2006 18:00:48 +1000
Roberto Michelena wrote:
ok, understood; so your FWA compensation is actually more like a
simulation of the FWA behaviour under a different illuminant.
Right, rather than the instrument giving a misleading measurement
of the color, it is compensated to give a reading as if the instrument
itself had measured the color under your intended illumination.
Usually when speaking of FWA compensation in commercial software
packages, it means aligning the measurements with human perception, in
other words simulating the effect of chromatic adaptation to the
substrate.
They don't do anything of the sort. They just (generally) us a fudge
to move the white point in a certain direction. Like using a UV
cuttoff filter, this may make your results better, or they
may make it worse. It depends on the situation.
I don't know what you mean by "simulating the effect of chromatic
adaptation to the substrate".
All packages simulate chromatic adaptation when they do conversion
of absolute readings to relative colorimetric. This has little to
do with FWA. When proofing though, one often uses absolute colorimetric,
so absolute accuracy of the color of the sample under the viewing
illuminant becomes important.
basically color management for proofing (assuming proofer gamut
completely contains reference gamut) can be reduced to three problems:
1) acquire reliable (low noise, repeatable) colorimetric or spectral
measurements of reference and proof eliminating influences from gloss,
translucency or other undesirable phenomena
2) generate a theoretical (read: numeric) metameric match between the
reference and the proof
3) ensure that those numbers do correlate with perception
#1 is instrument design and of course dependant on your proofer's
repeatability and noise levels
But also involves what assumptions the instrument is built on. Most
instruments assume that the light reflected at a given wavelength
only depends on the amount of illumination at that wavelength,
and the reflectance of that sample at that wavelength. Fluorescence
breaks this assumption. Argyll's FWA compensation repairs this break.
#2 is just the mathematic ability of the profiling algorithms and the
CMM; here there's debate between using a large patch number in a
single profile building step, or using iteration which effectively
amounts to a large patch number. And then also between using a PCS in
the conversion or using it only at the iterating stage but making the
conversion directly (link profile).
#3 is the real catch. Whereas #1 and #2 are totally objective and
measurable, #3 is where science still falls short. How many times
you've got a proof that verifies to less than 1dE from the reference,
and yet visually does not match. Different content of FWAs in proof
and reference is one common cause, and different illuminants are also
to blame.
The science doesn't necessarily fall short, it just hasn't been
applied most of the time. Most people don't bother to worry about
how their sample will appear in their actual viewing illuminant,
they just apply D50, and hope for the best. Similarly, most people
don't worry about how the viewing conditions will affect the visual
result. Most of the time, they don't worry about whether they're
viewing small areas of color or large, they just use the 1931
2 degree standard observer, and hope for the best. Most people
don't understand that individual observers vary in their color
perception, or that it changes as you grow older, they just
put up with disagreements amongst different people. And so it goes.
The fact is that the accuracy of most systems is not sufficient to
get perfect visual matches. Quite often I've seen matches of neutral
colors that were about 1 deltaE, and they looked different. But sure
enough, if difference could be reduced to something like 0.5 delta E or
better, it matched. Certainly the sampling is rather sparse to
get this sort of match across the whole gamut.
So the way you're doing FWA adaptation... well, it's correct in some
theoretical aspect but it's not what's usually needed. You're
modelling the spectra that will reach our eyes, but in doing so you're
failing to model the color that we'll perceive after chromatic
adaptation occurs.
I think you need to re-read your last sentence. I don't think it
makes a whole lot of sense. Modelling the spectra that reaches our
eyes is exactly what has to be done, to have a hope of modelling
how we see a color. If you don't have measurements to start with,
how can you go on ? I don't see that chromatic adaptation has
much to do with the subject at hand.
The FWA compensation in Argyll was developed specifically to improve
the visual match for proofing, and that's exactly what it does. We
proved it, time after time. You need to apply it correctly though.
Apply it incorrectly, and it will make things worse.
You can't just assume that because it predicts a b* = -11.3
under D50 for your output media, that it can't work. It can
and it does.
How much FWA does your source media have compared to your
destination media ?
What sort of light source are you viewing in ?
With FWA compensation off, how did your proofs look ?
Graeme Gill.
Other related posts:
