The normal human retina has four distinct populations of receptors: rods and three types of cone whose spectral sensitivities are shown in Fig. 8.2A. The curves are for a situation in which the same numbers of photons at each wavelength is delivered to the retinal receptors; the cones are labeled S, M, L for the simple reason that one is more sensitive to longer wavelengths, one to middle wavelengths, and one to shorter wavelengths. The wavelengths of maximal sensitivity of the different cone types are approximately 430, 530, and 560 nm. Figure 8.2B shows the same receptor functions, but now in terms of equal numbers of photons at each wavelength delivered to the cornea. Sensitivity to the shorter wavelengths seems greatly depressed. This is an artifact caused by the usual procedure in psychophysical studies of color vision in which the light stimulus is measured at the cornea: the ocular media, especially the lens, absorb many of the shorter wavelength photons before they reach the retina and so these lights seem less effective for vision; of course, one can allow for this by readjusting the number of photons delivered to the cornea (more on this later).

Because receptors are sensitive to some degree to the entire spectrum, it is misleading to refer to cones as red, green, or blue receptors; all receptors are simply light receptors. For wavelengths longer than about 520 nm, however,
the S-cones are so insensitive compared with L- and M-cones that these longer wavelengths effectively stimulate only the latter two types of cones; thus, anyone with normal color vision is a dichromat when the spectrum is limited to middle and long wavelengths. And, that is why a color-normal observer can make the match shown in Figure 8.1C: 590 nm = 550 nm + 630 nm, which is known as the Rayleigh match, enshrined in the Nagel anomaloscope. The necessary ratio of intensities of the 550 and 630nm lights is determined by the exact form of the L- and M-cone spectra (see Fig. 8.1C), which makes the anomaloscope a useful diagnostic tool; any significant deviation from the settings of the general population means that the observer has cones with unusual spectra. A genetic dichromat who lacks either M- or L-cones only has one photopigment in the spectral range of the anomaloscope; that person is effectively a monochromat over this range and so can match 590 nm with any ratio of 530 + 560 nm lights.

The Rayleigh match is deliberately limited to part of the spectrum because most color defects, especially of the genetic variety, involve only M- and L-cones. A person lacking L-cones is termed a protanope, one lacking M-cones is a deuteranope, and one lacking S-cones (a very rare person) is a tritanope. A normal adult has three classes of cones and is a trichromat; to match all possible lights, that individual needs to mix three primaries in the additive mixing field. Thus, merely making a normal Rayleigh match does not guarantee trichromacy; a tritanope, is a dichromat, but has the usual M- and L-cones and so would make normal Rayleigh matches.

Although we will continue to refer to three types of cone, it should be noted that this is not strictly correct. In fact, multiple forms of the genes exist that code for the L- and M-cone photopigments and many individuals express multiple forms (3,4). All individuals with normal color vision, however, still require only three primaries to match all other lights; regardless of how many different L- and M-cones we have, we are still rigorously trichromats, meaning that somehow the nervous system must combine signals from the variants of each cone type into signals that behave as if they came from single cone types (5).

Possession of different cones does not, by itself, guarantee color vision. There must also be a nervous system that compares the responses of the various cones to assess which responded more strongly and, thereby, gain information about the portion of the spectrum that stimulated the retina.

We start with retinal ganglion cells because they represent the output after all the processing at the level of the retina. Responses of
ganglion cells in the primate retina can be divided into two broad classes: spectrally opponent and spectrally nonopponent types. As with all ganglion cells in the higher mammals, they have circular receptive fields with concentric zones of excitation and inhibition; they are all spatially opponent, but what matters for color vision is which receptors drive these antagonistic responses. Two examples of the different organizations are shown in Fig. 8.3.

Both examples deal only with responses driven by L- and M-cones. Although these cones are more or less uniformly interspersed across each region of the retina, the responses from each of the receptive field’s concentric zones can be strongly biased toward one or the other cone type. Thus, for the spectrally opponent neuron in Fig. 8.3A, the center’s excitatory responses are driven by L-cones, whereas M-cones drive the inhibitory inputs that predominate in the surround. This neuron is reporting the net sum of its cone inputs (by convention, excitatory inputs are positive and inhibitory ones are negative). The result is that the neuron’s responses are spectrally opponent; a monochromatic light that covers the receptive field elicits one type of response for wavelengths to which the L-cone is the more sensitive, and the opposite sort of response when the M-cone is the more sensitive. Note, that a point in the spectrum exists to which this neuron does not respond, the point at which the L-and M-cones are equally sensitive so that their opposed inputs cancel. The location of this null point depends on the relative strengths of the L- and M-cone inputs to the particular cell.

By contrast, the neuron in Fig. 8.3B is spectrally nonopponent: both L-and M-cones contribute to responses of the excitatory center and the inhibitory surround so that any wavelength elicits the same type of response. In this example, it is assumed that L-cones are weighted more strongly and that the excitatory center is stronger than the surround, leading to the spectral responses shown.

The axons of the ganglion cells form the optic nerves that terminate in the lateral geniculate nucleus (LGN) of the thalamus. In primates (e.g., humans and macaque monkeys, species with very similar color vision), the LGN is a distinctively layered structure. The upper layers are termed parvocellular and the cell bodies of their neurons are small generically; they are called P-cells. The lower layers are magnocellular, cells that have larger bodies, called M-cells (beware of the distinction between M-cell and M-cone). A new category of very small cells, lying between some of the layers, has recently been described: these are termed koniocellular (K-cells) (6).

In macaques, all P-cells and some K-cells are spectrally opponent, whereas M-cells are spectrally nonopponent. All these cells have a spontaneous rate of firing action potentials in the absence of any light stimulus; excitation is signaled by an increase in firing rate from spontaneous, whereas inhibition is signaled by a decrease. The spectrally opponent cells can be divided into four subtypes, based on their spectral responses to monochromatic lights (7,8). Differences in the signs and relative weights of cone-derived inputs determine the particular responses: two (P-cells) are driven only by L- and M-cones, but with reversed signs (+L-M and +M-L), and the other two (K-cells) are driven by all three cone types (S vs. L+M). M-cells fall into two basic classes: some are excited to some degree by all wavelengths, whereas others are inhibited by all wavelengths; however, both types are driven only by L- and M-cones under light-adapted (photopic) conditions and by rods under dark-adapted (scotopic) conditions (9).

Normal human color vision is trichromatic, which means that it is defined by three independent dimensions, such as the intensities of the three primaries needed to match additively any other light. But this simply reduces color to this set of intensity axes, or some linear transformation of them as in a standard CIE chromaticity diagram (10). The matches specify when two lights appear the same, not what they appear like—red, pink, turquoise, and so on. Many possible trios of independent dimensions could be used. Sensations of color appearance are usually described using the three dimensions of hue (red, orange, yellow, and so on), saturation (quantity of hue: is it washed out and desaturated, or pure and saturated?), and brightness (in its everyday meaning of dim vs. bright).

How do we get from responses of neurons to such sensations? We must specify the linkages, the ways in which neuronal responses must be combined, presumably at higher levels of the central nervous system, to arrive at the sensations in which we are interested (11). Several things can be stated immediately: First, only the spectrally opponent cells can signal hue; when wavelength is changed, they are the only ones
whose responses change qualitatively, rather than just quantitatively. Second, the spectrally nonopponent cells respond in the same qualitative fashion to the entire spectrum, and so can only signal some intensive aspect of the stimulus (e.g., luminosity). Indeed, the nonopponent cells are probably the entire basis for the spectral sensitivity function; when the eye is light-adapted, the sensitivity of these cells matches the psychophysical photopic luminosity function (the particular function in Fig. 8.3B is the photopic luminosity function), and when dark-adapted, their sensitivity matches the scotopic luminosity function (9).

As far as hue is concerned, the four types of spectrally opponent LGN neurons were originally thought to underlie the four basic hue sensations of red (R), yellow (Y), green (G), and blue (B) (7). That early hypothesis, however, was incomplete. It is now clear that the responses of the LGN cells do not exactly match color appearance functions, such as those obtained from direct hue-scaling psychophysics. The LGN opponent cells, however, are the necessary inputs to higher cortical levels at which they are compared with each other and recombined to produce mechanisms that directly specify hue sensations (5,8,12, 13).

The cortical recombination of spectrally opponent LGN responses produces two generic hue mechanisms: +R-G and +Y-B (the signs are arbitrary and serve only to distinguish qualitatively opposed responses). In this model, when the +R-G hue mechanism is excited, the sensation includes red; the exact sensation depends on the simultaneous activity in the other hue mechanism, +Y-B. If both are excited, the sensation will include some red and some yellow, the precise ratio depending on the degree to which each mechanism is excited. Note that the hue mechanisms have spectral null points, at which their opposed inputs cancel each other; for example, the +R-G mechanism has such a point at approximately 575 nm. At that point, the sensation includes neither red nor green, but the other mechanism, +Y-B, is strongly excited, leading to a sensation of unique yellow. Similar comparisons apply to all the other points in the spectrum; across the spectrum, hue shades continuously from R through RY, to Y, GY, G, BG, B, and finally BR.

The final sensory dimension of saturation probably involves a comparison of responses of spectrally opponent and nonopponent neurons. For example, if only the luminosity-signaling nonopponent cells respond to some stimulus, then a sensation of light would occur, but with no hue—the stimulus would have elicited a sensation that was completely desaturated. As soon as one or more of the opponent neurons also responds, a sensation of some hue must occur, whose saturation would be determined by the total hue-signaling responses relative to all the responses elicited by the stimulus.

A genetic dichromat cannot have the same range and variation of hue sensations for the simple reason that the opponent neuronal channels can only compare responses of two cone types. One consequence is that there will be some narrow portion of the spectrum in which the responses of these two cones are equal to each other; at that point, the responses of the opponent channels will cancel each other, eliminating any sensation of hue. The only sensation at that point will be caused by nonopponent channels that will produce a sensation that is completely desaturated. Each dichromat has a spectral neutral point. The spectral region of this neutral point varies with the type of dichromacy; protanopes and deuteranopes have neutral points in the vicinity of 490 to 510 nm, with those of deuteranopes being slightly longer; tritanopes have their neutral point in the vicinity of 465 to 475 nm.

None of the above methods is “fool-proof.” Ultimately the caveats found in virtually every paper can only be resolved by convergent lines of evidence from different studies, using different methods, and making different assumptions.