Color - Computer Graphics: Principles and Practice - page 769

primaries X , Y , and Z ; if a color C is generated by X X + Y Y + Z Z , we can think
of X , Y , and Z as “coordinates” for that color in the space of all possible colors.
There are other coordinate systems on the space of colors, such as the CIE
L ∗ u ∗ v ∗ and L ∗ a ∗ b ∗ , in which L ∗ captures the notion of intensity, while the other
coordinates encode chromaticity. In these systems, distances between color triples
correspond to perceptual distances much more closely than do distances between
( X , Y , Z ) -coordinate triples. But the coordinates in these systems are not linear
functions of the X , Y , Z -coordinates (which are in turn linear functions of radio-
metric quantities), so they are not suitable for computations with a physical basis.
In both of the “perceptual” coordinate systems, there's a free parameter,
namely, the color chosen as “white.” Without the knowledge of the white point,
you cannot convert an L ∗ u ∗ v ∗ coordinate triple into an XYZ triple, for instance.
We now move on from the description of color to the question of how to rep-
resent color in an image file, a television signal, etc. Considering that there's only
a half-century of experience in this regard, a surprisingly large number of repre-
sentation methods have arisen.
28.11 White
As we mentioned earlier, many spectral power distributions appear white, so pick-
ing a particular white point can be a challenge. And an SPD that looks white at
one intensity may look yellow at another intensity, because of the adaptation of the
eye. Furthermore, the surroundings may have a substantial impact on the appear-
ance of a color; if we watch a slide show in a dark room, showing a scene illumi-
nated by incandescent lamps, we rapidly accommodate so that the white point of
the slides appears white. But if that same slide show is shown in a well-lit room
with white walls, the “white” within the slides may appear yellow, for instance.
The CIE has defined several standard “whites”; the simplest (from the point
of view of computation) is illuminant E , which has a constant SPD across the
range of visible light. Illuminant C , now deprecated but still widely used, attempts
to approximate the white of sunlight. More common in modern usage are the D
series of illuminants, which are tabulated by the CIE in 5 nm increments. Many
of the most useful are, at a gross level, quite similar to blackbody distributions,
and the names indicate this: D65 is similar to 6500 K black body radiation, D50 is
similar to 5000 K radiation, etc. The photography industry uses the D55 standard;
either this or D65 is a good choice for much of computer graphics.
28.12 Encoding of Intensity, Exponents,
and Gamma Correction
As mentioned above, the CIE standard for defining L ∗ uses a 3 -power law; the idea
is that L ∗ is a reasonable measure of perceived brightness of light (at least within
a modest range of luminances around the luminance of some reference white).
Suppose that you wanted to store or transmit information about light without using
too many bits. If you were engaged in physical measurements, you'd just want to
choose some numeric representation of intensity. But if you were planning to use
the information about light in some way that involved a human looking at it (e.g.,
if you were a television engineer trying to decide what information to encode in