1. Opponent colour theory
2. Electrophysiological studies of opponent processes
3. Psychophysical studies of opponent processes
4. Luminance
5. An analogy with colour television
6. Cortical analysis of colour
1. Opponent colour theory
Historically, a number of rivals to the trichromatic theory have taken as their starting point the phenomenology of colour. The most celebrated of these rivals is the 'opponent colour' or 'tetrachromatic' theory of Ewald Hering. The trichromatic theory, in the form that it was advanced by Thomas Young, James Clerk Maxwell, and Hermann von Helmholtz, does not obviously account for the fact that a mixture of yellowy red and yellowy green can produce a yellow that is without trace of redness or greenness. Hering proposed that colour analysis depends on the action of two types of detector, each having two modes of response: one detector that signals either red or green, and a second that signals either yellow or blue. Finally he supposed that brightness depends on a system that signals white or black. Thus one can see a colour that looks greenish, and bluish, and dark, but not one that is both reddish and greenish, or both light and dark, and so on.Nowadays it is generally held that a complete theory of colour vision must draw elements from both the trichromatic and the opponent colour theories. Indeed, Helmholtz himself proved that the two theories are not incompatible, since a simple transformation could change the three receptor outputs into two difference signals and one additive signal. Wasserman (1978) gives Helmholtz's argument. Unfortunately, this statement by Helmholtz appeared only in the second edition of his Handbuch der physiologischen Optik (1896) and was missing from the third edition (1909), the latter being the one translated into English (by the Optical Society of America, in 1924).
2. Electrophysiological studies of opponent processes
A transformation of the kind considered by Helmholtz probably occurs in the retina itself. In 1958, MacNichol and Svaetichin found that, at the level of the horizontal cells in the goldfish retina, light of one wavelength may depolarize (i.e. excite) a cell, whereas light from a different part of the spectrum may hyperpolarize (i.e. inhibit the cell). In the absence of stimulation the cell's electrical potential has an intermediate value, and this potential can be increased or decreased by stimulating the retina with different colours. Soon afterwards De Valois reported opponent responses in the lateral geniculate nuclei of macaque monkeys, and others have reported such responses at the level of retinal ganglion cells in the monkey, although it has so far proved difficult to obtain recordings from the more distal cells of the primate retina. Thus De Valois was led to propose that colour information is transformed from a 'component' system (three cone types giving responses that increase with increasing intensity of stimulation) to an 'opponent' system in which some wavelengths cause an increase in a given retinal signal and other wavelengths cause a decrease. Such an opponent system effectively transmits colour difference signals: for example, a cell that increases its firing rate as a result of stimulation with red (R) light and decreases its firing rate with green (G) stimulation can be said to be signalling + R − G.If colour difference signals are transmitted, can we tell which combinations of receptor responses are 'differenced'? There is some doubt about this, since the electrophysiological reports are not entirely consistent, but by and large, if we call the three cone types R, G, and B, neural pathways have been found that signal ± (R − G) and ± [(R+G) − B]. It should be noted that in the second of these expressions, the sum (R+G) may be taken as signalling 'yellow'. This may be the explanation for why yellow is the fourth psychological primary — along with red, green, and blue. It seems that the responses from the R and G cones are added, thus generating a spare colour (yellow) that looks neither green nor red, and is complementary to blue. For details of the physiological evidence for opponent processing, see De Valois and De Valois (1975).
3. Psychophysical studies of opponent processes
There is good psychophysical evidence that opponent colour channels exist. Thus the efficacy of two adapting fields of different wavelength (or the detectability of two lights of different wavelength) may be less when the two are concurrently presented than when either alone is present (see Mollon 1982). In these experiments the observer is asked only to detect a liminal (difference) stimulus; he is not asked to make subjective judgements about the quality of the colour appearance.It is currently a controversial question how closely these psychophysically demonstrated opponencies are related to the phenomenological oppositions of red and green and of yellow and blue. The latter oppositions, however, can be systematically measured. This was classically done in the 1950s by Jameson and Hurvich (see Hurvich 1978), using the method of hue cancellation: to measure, say, the amount of yellowness in a series of wavelengths from the long-wavelength part of the spectrum, the experimenter adds blue light of a fixed wavelength to each in turn of the long wavelengths. The long wavelengths would range from a slightly yellowish green to a deep red. For each of these test wavelengths, the observer is asked to adjust the added blue light until the mixture looks neither yellowish nor bluish. The amount of blue light required is the dependent variable and is taken as a measure of the yellowness, the 'yellow chromatic valence', of the test wavelength.
4. Luminance
The main tenet of opponent colour theory is that there are two independent 'channels' signalling colour information. One of these signals red or green, the other signals yellow or blue. But from trichromatic theory it is clear that we need three, not two, independent variables to describe colour appearance. Thus, a third channel is necessary. What does this channel signal? The answer is luminance. The signals from the R and G (but probably not B) cones are added, and the information is transmitted as luminance. But again, it is opponent coded. The channel signals 'brighter' or 'darker', but with respect to what? Simply with respect to other parts of the picture, or retinal image. The luminance channel signals comparative information, not about absolute local luminance but about how the local luminance compares with the rest of the scene. Comparisons are made not across colour but across space. The visual system places heavy emphasis on comparison across space. The impressive colour effects demonstrated by Edwin Land (see retinex theory and colour constancy) are a powerful example of long-range interaction in the visual system.5. An analogy with colour television
Why should opponent coding of colour have evolved? It may be useful to draw analogies with colour television. About 1950, television engineers started looking for a means of transmitting an acceptable colour picture efficiently. This meant transmitting as little information as possible (i.e. keeping the bandwidth low), with a high level of resistance to extraneous noise, and in a way that was compatible with monochrome receivers, i.e. capable of giving an acceptable black-and-white picture.Obviously this analogy cannot be pursued too far, since colour television aims to provide an image that is a good substitute for the original object; the visual system of each member of the audience must then get to work on the picture. However, the choice of how much importance to attach to different aspects of the information contained in the picture, and how to transmit the information most efficiently, seems to raise interesting questions for visual science.
The simplest system would 'look at' the original scene through three primary filters, transmit the three pictures separately, and recombine them in the receiver. Since the G primary is most similar to the luminance signal, this would be the one that monochrome receivers would receive. But errors would result, reds being too dim and greens too bright. More importantly, three times as much information would have to be transmitted (so the bandwidth would be trebled) in comparison with monochrome transmission. This would be expensive, and impossible where many broadcasts must share a limited range of radio frequencies.
At the time when colour television was being developed, it was known that visual acuity was worse for colour than for brightness contrast. Since it was desirable to separate colour and luminance anyway (to allow monochrome sets to receive a true luminance signal) there now appeared a way of reducing the total transmission bandwidth: give the colour information less bandwidth than the luminance information. But how could colour and luminance be separated? A neat mathematical solution presented itself: opponent coding. If R, G, and B are the signals from the red, green, and blue receivers respectively, then we can define luminance, L, as
| L = 0.3R + 0.59G + 0.11B | (1) |
We can now derive two 'colour difference equations':
| R − L = 0.7R − 0.59G − 0.11B | (2) |
| B − L = −0.3 R −0.59 G + 0.89 B | (3) |
There are other consequences of the colour difference operation that hold advantages for both television and the visual system. If three trichromatic signals were transmitted untransformed, then redundant information would be carried in the transmission. For the three signals would be correlated: they would necessarily be correlated because the spectral sensitivities of the three receptors overlap and because objects in the world have broad spectral reflectances. To avoid waste of valuable channel capacity the three signals should be orthogonal, i.e. should not be correlated with each other. By transmitting a luminance signal and two difference signals, visual systems may approach this ideal of communications engineering (Fukurotani 1982).
Another advantage of the difference operation may lie in the preservation of neutral colours. Imagine that a television camera is looking at a neutral object. There will be a luminance signal, but the two chrominance signals will be zero, since the camera is set so that R = G = B. The chrominance signals are thus zero: all is well. But now imagine that the gains of the two chrominance channels are altered — that is, the R − L and G − L signals are each multiplied by a different number (as might arise from drift in the electronics). The two channels will still be signalling 'zero' for neutral scenes, since a product of any number and zero is still zero. The resistance of the neutrality of neutral objects to drifts in the gains of opponent channels is likely to be an important consideration in our visual system as well.
6. Cortical analysis of colour
A question of great interest, and still unsettled, is that of the extent to which colour is analysed separately from other attributes of the retinal image, such as form and movement. S. Zeki has identified in the rhesus monkey two adjacent regions of the prestriate cortex that he suggests are specialized for the analysis of colour: these regions (denoted 'V4' and 'V4A') lie on the posterior bank of the lunate sulcus and the anterior bank of the superior temporal sulcus (Zeki 1977). He suggests that colour-specific cells are here more frequent than in other prestriate regions and, most interestingly, that the cells show colour constancy when one patch of a complex, multicoloured array falls within their receptive field; that is to say, they respond to the colour seen by a human observer despite large changes in the local spectral flux that falls on their receptive field. But as yet we have little idea of how the colour constancy is achieved or of how the hue of an object is referred to its other attributes, such as movement, shape, and distance.(Published 1987)
— Tom Troscianko
- Bibliography
- De Valois, R. L., and De Valois, K. K. (1975). 'Neural coding of color'. In Carterette, E. C., and Friedman, M. P. (eds.), Handbook of Perception, vol. v.
- Fukurotani, K. (1982). 'Color information coding of horizontal-cell responses in fish retina'. Color Research and Application, 7.
- Hurvich, L. M. (1978). 'Two decades of opponent processes'. Colour, 77.
- Mollon, J. D. (1982). 'Color vision'. Annual Reviews of Psychology, 33.
- Sims, H. V. (1969). Principles of PAL Colour Television and Related Systems.
- Wasserman, G. S. (1978). Color Vision: An Historical Introduction.
- Zeki, S. M. (1977). 'Colour coding in the superior temporal sulcus of the rhesus monkey visual cortex'. Proceedings of the Royal Society Series B, Biological Sciences, 197.
