Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!cwjcc!gatech!bloom-beacon!dawn.UUCP!stpeters
From: stpeters@dawn.UUCP
Newsgroups: comp.windows.x
Subject: Re: Luminance from RGB
Message-ID: <8811132229.AA02726@dawn.steinmetz.GE.COM>
Date: 13 Nov 88 22:29:31 GMT
Sender: daemon@bloom-beacon.MIT.EDU
Organization: The Internet
Lines: 77

pixar!mab@bloom-beacon.mit.edu  (Malcolm Blanchard) writes:
> The digital color correction was abandoned and the production
> relied on the film lab to produce an aesthetic balance.  Thus proving
> to me that science may work, but computer graphics and film making are
> still largely a matter of art.

Amen.

I am weary of claims that the CIE description of 1931, as pivotal as
it was, is the last word in human color perception.  Does it really
suprise anybody that work over the last half century has unearthed
some aspects that the CIE did not?

CIE is a comfort to engineers, because it wraps color in a widely
accepted veneer of quantification, making it seem like a science,
something they can deal with.  It allows them to talk about color in
terms of "CIE coordinates", quoted to two or more significant figures,
as for phosphors in a recent posting.

However, the CIE coordinates are strictly applicable only to truly
continuous spectra.  Phosphors emit much of their output in a few
narrow spectral lines.  Further, the proportions of the emission among
the peaks - and the continuous underlying background - depend on the
excitation.  Assigning precise CIE coordinates to such phosphors is
meaningless.  (Note the word "precise".)

Human vision receives input in four channels.  Three have spectral
response curves that peak in red, green, and blue parts of the
spectrum respectively, but each responds somewhat throughout most of
the visible spectrum.  The fourth channel is the roughly colorless
night vision channel, responding to overall brightness.

Our visual system appears to assign color based on ratios of the
inputs - and on its own biases.  It is an ornery system indeed and is
quite willing to reject input that does not fit: when two images of a
scene, each projected through a different narrowband red filter, are
superimposed on a screen, it is quite possible for humans to perceive
greens and blues in a scene, in spite of information from the eye's
receptors that the overwhelming majority of received light energy is
in the red part of the spectrum[1].

As far as I know, nobody knows why for sure.  However, apparently our
visual system (evolved for broad-spectrum illumination) has more
confidence in the differing responses of the blue, green and
"colorless" receptors to the two red peaks than in the evidence from
the single red channel.  With inputs voting 3:1 in favor of other
colors, it ignores the red input.

If you had a red phosphor that emitted primarily in two spectral peaks
and the proportional distribution between the peaks depended on the
excitation, you could perceive the emission from the red phosphor as
being blue or green at some intensities.  (The emission would have to
be confined virtually entirely to the peaks, whereas real phosphors
emit a continuous background as well as peaks, so such a phosphor
would be very unlikely.)

Not only does the concentration by phosphors of their output into a
few peaks distort the color response they generate from their nominal
CIE coordinates, but the spectrum of each phosphor interacts with that
of the others in an RGB system, probably by altering response ratios.

Two projects ago, I spent several years simulating color television
systems, from the camera to NTSC baseband to RF to NTSC baseband to
final "RGB", as well as simulating high-definition TV schemes.  I know
all too well how hard it is to give up the comfort of quantifying
color.  CIE was a good start, a necessary and useful approximation for
technological progress like color TV transmission and color CRT
development.  However, "accurate" color is still an art.

--
Dick St.Peters                        
GE Corporate R&D, Schenectady, NY
stpeters@ge-crd.arpa              
uunet!steinmetz!stpeters

1. The experiment requires that each (B&W) slide be projected through
   the same filter through which it was photographed.