Xref: utzoo comp.graphics:3562 comp.windows.x:6000
Path: utzoo!attcan!uunet!seismo!sundc!pitstop!sun!vector!poynton
From: poynton%vector@Sun.COM (Charles Poynton)
Newsgroups: comp.graphics,comp.windows.x
Subject: Luminance from RGB
Summary: If (R+G+B)/3 is white, what happens when "blue" shifts to UV?
Message-ID: <76649@sun.uucp>
Date: 9 Nov 88 02:00:33 GMT
References: <8811042303.AA21505@dawn.steinmetz.GE.COM>
  <8811070533.AA08904@vector.sun.com>
  <8811080030.AA14681@EXPO.LCS.MIT.EDU>
Sender: news@sun.uucp
Lines: 89

This follows Dick St Peters' comment <8811080030.AA14681@EXPO.LCS.MIT.EDU>
of Comp.windows.x that (R+G+B)/3 is an appropriate weighting function for
luminance.

Summary

Luminance weighting coefficients are a function ONLY of the human
luminance sensitivity functions, the selected phosphor chromaticities, and
the chromaticity of the reference illuminant (white point).  The equations
can be found on page 2.28 and on in K. Blair Benson's "Television
Engineering Handbook", or on page 20-10 of Donald G. Fink's Electronic
Engineers' Handbook, Second Edition".

A Simple Thought Experiment

Think of three monochromatic primaries, for example, tuneable dye lasers.
If (R+G+B)/3 is white, what happens when I move the wavelength of the blue
primary?  If I move it towards the ultraviolet, the eye responds less to
it and luminance drops.  If I move it towards green, the eye responds more
(in which channel(s) we don't care) and its luminance increases.

Hence, the equation for luminance must depend in some manner on what
wavelength you call "blue".

White

CIE Illuminant D65 refers to a standard spectral distribution which is
chosen to approximate daylight.  This is television's standard white
reference.  So-called "Zero chroma" in NTSC occurs on this colour.  People
may or may not respond to this as "white":  if the observer is adapted to
a pink room, he'll think it's yellowish-green.  The reason that the CIE
decided to standardize white, in 1931, is to allow work in colourimetry to
be done objectively.

To achieve accurate colour reproduction requires specification of
reference white.  This is a lurking problem for accurate colour
reproduction in computer graphics, because most workstation monitors are
adjusted for a white point (of about 9300 K) that's quite a bit more blue
than the television standard.  Software people tend to say
"R=G=B=1=white.  Simple."  But it's not that simple.

Blue Shirts

Sorry to air the laundry in public.  Fabric whiteners work by adding
materials which flouresce to convert ultraviolet light [to which the human
eye is not sensitive] into visible light in the blue part of the spectrum
[to which it is].  The shirt not only looks brighter than white, it IS
brighter than white.  But this has nothing to do with the luminance
coefficients.

With modern phosphors the blue luminances coefficient is actually tending
to go DOWN.  To produce accurate reproduction with European standard
phosphors requires a blue luminance contribution of 0.071.  This is a big
colour problem for Europe, because they use the same luminance function as
we do, and it's not exactly matched to their phosphors.  They tweak their
camera spectral sensitivities as a first-order fix to improve the colour
reproduction accuracy.  

Orthogonality

Luminance is luminance; chrominance is chrominance; and NTSC can be
thought of as conveying Y, U, and V independently.  This is true
regardless of your interpretation of what these quantities represent.
They're "orthogonal" provided that one can't be derived from the other
two.  Although there are some subtle signal-to-noise ratio considerations
in television coding, this issue is independent of (or should I say
orthogonal to) the choice of luminance coefficients.

NTSC RF Modulation

I stand corrected by Mr St Peters on an RF modulation point:  there IS one
compromise made in NTSC transmission via RF.  White modulates the RF
carrier to 12.5%, and black modulates it to 70.3125%.  When NTSC modulates
an RF carrier, chroma excursions near highly saturated yellow and near
highly saturated cyan are clipped to 120 IEEE units prior to the
modulator, to avoid UNDER-modulating the transmitter.  Two small regions
of colour space are lost in this case.  No practical television camera has
sufficient colour separation capability to generate signals in these
regions, but electronically-synthesized colour bars have perfect colour
saturation and would undermodulate the transmitter if left alone.  Studio
colour bars are generated at 75% saturation to avoid these two regions of
colour space.  This issue does not bear on the choice of luminance
coefficients, and is relevant to broadcast only.  VHS machines and NTSC
monitors reproduce all of the NTSC colour space even at 100% saturation.

I would have put this first, but then you wouldn't have read all the
colour stuff, would you?

Charles