Previous: Basic Color Terms Up: Related Work Next: Color Models

Neurophysiology of Color Vision

There is a considerable body of literature on the neurophysiology of color vision, some references to which can be found in [Boynton 1979], [Boynton 1990], [Dow 1990], and [Hubel \& Livingstone 1990], for instance.

The work of De Valois and his colleagues is some of the best known in this area [De Valois \& De Valois 1975][De Valois \& Jacobs 1968][De Valois et al. 1966]. They made extracellular electrical recordings of LGN cell responses in the Macaque monkey, while the visual system was stimulated with monochromatic light. They cite various kinds of evidence that indicate that the Macaque's vision system is very similar to that of humans, and conclude that it is safe to generalize the experimental results to human neurophysiology. They identify six types of cells, based on a statistical analysis of the responses of a large population of cells. Four of these cell types display what they call spectrally opponent responses, and two types display spectrally non-opponent responses. The opponent cell types respond with inhibition to one part of the spectrum, and with excitation to another part. The non-opponent cell types respond with either inhibition or excitation to the entire spectrum. One can further distinguish three pairs within the cell types; within each pair, the members display ``mirror imaged'' responses: when one type displays inhibition in response to some part of the spectrum, the other displays excitation, and vice versa. De Valois et al. note that the response functions of the six cell types seem to agree well with various psychophysical findings on color perception, and are reminiscent of the opponent color response functions of [Hurvich \& Jameson 1957]. I will discuss these findings in greater detail in Chapter .

The existence of color opponent cells in the (Macaque) LGN has been confirmed in numerous studies, for instance [Derrington et al. 1984]. The latter study finds groups of parvo-cellular (P) cells that get opposed but not equally balanced inputs from only ``red'' and ``green'' cones, which they refer to as ``R-G'' cells, and different groups of P-cells that receive inputs from ``blue'' cones almost equally opposed by some (varying) combined input from ``red'' and ``green'' cones, which they refer to as ``B-(R&G)'' cells. In the magno-cellular (M) layer of the LGN they find cells that are both spatially and chromatically opponent, where the ``red'' and ``green'' cones contribute differentially to the center and surround of their receptive fields. Some of these cells have inputs from ``blue'' cones as well.

The work presented in [Hubel \& Livingstone 1990] discusses more LGN cell types than [De Valois et al. 1966] does, in particular, the different characteristics of parvo-cellular vs. magno-cellular responses. It also uses a different classification scheme, which has become widely accepted after the work of De Valois et al. was published. Some of these cell types are more sensitive to luminance contrast, others to chromatic contrast, and there are various kinds of receptive field organizations. These characteristics are not neatly separable, and it seems that for any given response dimension (like luminance or chromatic contrast) there is a range of differently sensitive cells. This organization is reminiscent of the orientation columns in the cortex that consist of cells tuned to a particular direction of edges in their receptive field, with the range of possible directions sampled at approximately regular intervals by differently sensitive cell types [Dow 1990]. In addition, there are differences in temporal response characteristics across cell types. The authors report similar response characteristics and ranges for cortical cells, in addition to the presence of motion-selective cells and other types. They also present some further evidence that the Macaque's visual system is nearly identical to the human one in terms of color perception and other dimensions.

There are no real spectral sensitivity measurements in [Hubel \& Livingstone 1990], as there are in [De Valois et al. 1966], and the kind of stimuli used is different. For the purpose of classifying different cell types in detail, and tracing the pathways between the LGN and the visual cortex, the kind of stimuli used by [De Valois et al. 1966] may not be the most appropriate, but for the purpose of measuring the spectral response of groups of similar cells, they probably are. I will base what follows on the results reported in [De Valois et al. 1966], while realizing that this isolates a particular response dimension in a way that may not be fully supported by neurophysiological data. I feel this simplification is justified though, at least at this stage of modeling, since my work attempts to deal with color only, as a self-contained (set of) perceptual variable(s). It is in this context that I interpret Boynton's warning:

The facile identification of signals recorded from cells in the LGN with various aspects of color appearance, which is now deeply rooted in many introductory textbooks, is both premature and misleading. [Boynton 1990][p. 228]
I will use these LGN recordings as a starting point to construct a color space, which in turn will give rise to measures that I will relate to color appearance, but this is by no means a ``facile identification''.

lammens@cs.buffalo.edu