Previous: CIE chromaticity and related models
Up: Color Models
Next: Computer graphics and computer vision color models
As [Boynton 1990] points out, it has been generally understood for
nearly 200 years that the initial basis for color vision lies in the
differential excitation of three different classes of cone photoreceptors
in the retina (Figure ; see also
Section
).
If we represent a stimulus as a spectral power distribution (SPD), which is
in turn represented as a function of wavelength , as are the cone
sensitivity functions
,
, and
,
and if we further assume that the response of the visual system is linear,
we can represent the response of each of the cone types to the stimulus as
follows (Section
):
The entire SPD is thus reduced to three numbers only. Obviously there is
considerable loss of information in this transformation, so that many
different SPDs will result in the same set of three LMS values. Any two
stimuli that result in identical LMS values cannot be distinguished from
each other; we call such stimuli metameric (Figure ). This
reduction can be seen as a first phase of categorization. The next phase,
color perception and naming, occurs after the transformation to three
values, and is the main topic of this dissertation. Without serious ``data
reduction'' and categorization, the world would be a place of ``blooming,
buzzing confusion'', as William James put it.
It is the phenomenon of metamerism, based on three different receptor sensitivities, that has made color TV and color computer graphics possible. If our visual system were sensitive to the shape of SPDs themselves, i.e., to differences in the amplitude of individual wavelengths, rather than to the weighted energy content of three broad spectral bands, it would be much harder to reproduce color in any way.
In the cases of color TV or color vision research, a scene is captured using
photosensors with broad spectral sensitivities resembling those of human
cones (Figure ) [Ballard \& Brown 1982][McIlwain \& Dean 1956].
Once an image is captured, colors are defined as RGB triplets. In computer graphics work, these RGB triplets are generated directly, e.g., from an underlying imaging model [Hill 1990][Foley \& Van Dam 1982]. The colors are then reproduced by using the RGB values to drive three independent electron guns in a cathode ray tube, each activating a particular kind of electroluminescent phosphor. The combined light emitted from the three phosphors will then ideally result in cone excitations that are proportional to what they would have been if the scene were viewed directly. RGB values can also be used in color printing, after conversion to the complementary CMY (Cyan, Magenta, Yellow) colors, using subtractive rather than additive principles [Rogers 1985].