Opponent Process Theories of Color Vision
Ratios instead of summations
In a previous post, I discussed Trichromatic Theories of Color Vision that were originally formulated by Young and Helmholtz in the 1800s.
https://www.cantorsparadise.com/trichromatic-color-theory-33f1eefbaa68
These theories explain color matches based on linear summations of photon absorptions in three different kinds of cone photoreceptors. In this post, I am going to describe a different approach to understanding color perception, Opponent Process Color Theories. The origin of these theories is attributed to Ewald Hering in 1892. They employ ratios to explain certain aspects of color perception.
For several decades stretching into the 20th Century, the Trichromatic and Opponent Processes were considered to be competing for theories and treated as though the scientific goal was to figure out which one was correct. It is now understood that the two theories simply apply to different aspects of color perception, and can be related to the ways color information is coded at various stages of neural processing. Trichromatic theories are mostly concerned with trying to explain how certain mixtures of wavelengths can be made to match each other in appearance. Opponent theories focus on trying to understand the finding that certain kinds of color experiences appear to operate in opposition to one another.
One example of color phenomena that prove difficult to understand within the scope of trichromatic theory but are more amenable to the explanation from the perspective of an opponent-process theory is color adaptation aftereffects. You can experience one of these yourself as follows. First stare at the circle in the top panel of the following figure for 30 seconds or more, keeping your eyes focused on the crosshair at the center. Then quickly change your gaze to focus on the crosshair at the center of the gray panel immediately below it.
Most humans with normal color vision report that a color afterimage appears over the gray square for at least a fleeting period that can last up to several seconds. If you have normal color vision and see a color afterimage,¹ pay attention to the colors experienced. In the adapting target (top panel), starting at the 12 o’clock position and moving clockwise, the colors are: green, yellow, blue, red. In the afterimage, the corresponding colors that are typically seen are: red, blue, yellow, green.
Opponent Process theories postulate that color information is transmitted by three channels. One transmits either green or red but cannot do both at the same time. Similarly, a second channel transmits blue or yellow. A third carry black or white. For purposes of this discussion, we will ignore the third channel, except to note that both trichromatic and opponent-process theories posit three channels. The channels are simply organized differently. The channels that I described in my earlier post about trichromatic theories correspond to three cone photoreceptor types that respond maximally to short (“blue”), middle (“green”), and long (“red”) wavelengths. The three channels proposed by opponent-process theories are G/R (“green/red”), B/Y (“blue/yellow”), and B/W (“black/white”).
An oversimplified, but convenient, explanation of color adaptation aftereffects are as follows. Consider the green quadrant shown in the upper panel of the demonstration. The portion of the retina that is stimulated by this quadrant is activating the G/R channels such that they signal a sensation of “green.” After viewing the figure for an extended period the G components in these channels become somewhat fatigued. When the gaze is moved to the lower panel, the gray field is providing equal amounts of stimulation to the G and R components. However, since the G components are fatigued, they will produce a weaker response than the R, producing weak red sensations for a brief period until the channel has recovered to its normal balanced state.
There are a number of other color phenomena, in addition to adaptation effects, that are more easily explained by opponent-process than trichromatic theories. Another example comes from experiments in which observers are asked to apply color names while viewing mixtures of wavelengths. Observers will sometimes report, for example, that a mixture appears to have tinges of both “red” and “yellow”, but do not usually report seeing tinges of both “red” and “green” at the same time. This can be explained by the fact that G/R and B/Y can both be transmitting at the same time (producing simultaneous red and yellow sensations), but the G/R channel can only transmit “red” or “green” but not both at the same time.
The modern understanding of opponent-process theories is grounded in neuroscience studies of neural processing in the visual pathways. I am going to greatly simplify how neural processing in the retina actually takes place here in order to explain how opponent-process neural channels work without getting too bogged down in anatomical and physiological minutia. I am going to describe a few general properties of neurons called retinal ganglion cells. The axons of these neurons transmit information up the optic nerve from the retina to the brain. These neurons exhibit spontaneous activity, meaning they are actively firing neural impulses all the time, whether receiving visual stimulation or not. When they receive excitatory inputs they increase their firing rate above the spontaneous level (positive influence on the output) and similarly decrease firing when inhibitory inputs are received (negative influence on the output). Retinal ganglion cells do not actually receive inputs directly from cone photoreceptors because there is processing by intermediate neurons located within the retina before cone signals reach the retinal ganglion cells. However, for purposes of this discussion, those details are not important, and we will consider the neural activity of a hypothetical retinal ganglion cell as though it receives direct excitatory input from receptor type A and inhibitory input from receptor type B as illustrated schematically here.
Photoreceptors A and B each simply sum the number of photons absorbed. However, the neural signals output from the receptors that feed onto the retinal ganglion cell is not linearly proportional to the number of photons absorbed. There is a compressive nonlinearity such that the inputs from the receptors to the retinal ganglion cell can be characterized, at least roughly, as the logarithm of photon absorptions. As a result, the A minus B output of the retinal ganglion cell is related to the ratio of photon absorptions in the two receptors:
Output = log (photon absorptions in A) — log (photon absorptions in B)
= (photon absorptions in A) / (photon absorptions in B)
The following diagram shows the spectral absorption curves for these two hypothetical photoreceptor types, A and B.
Note that each wavelength in the visible spectrum will produce a unique ratio.² In other words, there is a unique color code for every wavelength. Mixtures of wavelengths will also create some particular ratio between the outputs of A and B, and the color code for that mixture will match the color of the single wavelength having that same ratio.
Consider the point where the two curves cross, designated by the thick arrow in the diagram. The ratio associated with that wavelength is 1/1 and is referred to as the neutral point for the A/B channel. The color designated by the A/B channel at this neutral point is gray, or more accurately, achromatic. This code is transmitted when no light is present, allowing the hypothetical retinal ganglion neuron to fire at its spontaneous rate. The same code is transmitted when a single wavelength at the neutral point is present. Finally, it is also carried when white light (consisting of equal amounts of light at all wavelengths across the visible spectrum) is present.
If an observer had only two receptor types as illustrated in this A and B example, there would, necessarily,³ have to be present some single wavelength (the neutral point) that could not be discriminated from white light. This situation applies to some forms of human color blindness (various forms of dichromacy) and provides the rationale for a common way of diagnosing these conditions. The individual is tested with a series of wavelengths to see if a neutral point can be found that cannot be discriminated from white light.
However, a human with normal color vision has three receptor types so the only way a single wavelength would mimic white light would be if all three absorption spectra crossed at the same wavelength, a highly unlikely possibility as illustrated in the next figure.
Cone photoreceptors having the three absorption spectra labeled as A, B, and C in this hypothetical example are actually present in humans with normal color vision. They are given different labels in different contexts. In trichromatic color theory, they are referred to as Short (S), Middle (M), and Long (L), designating the wavelengths where their maximum absorptions take place. They are also sometimes referred to, using labels that are more confusing than helpful, as the Blue, Green, and Red photoreceptors, based on the color one would perceive while viewing a single wavelength located at the peak of each of their spectral absorption curves. This somewhat confusing terminology has been carried over to opponent-process theories when labeling the opponent channels. The G/R channels get opposing input from the “Green” and “Red” receptors. The B/Y channel gets one input from the “Blue” receptors and an opposing input from a combination of the “Green” and “Red” receptors. Since a mixture of “green” and “red” light looks “yellow”, this combination is given the label Y in the B/Y channel. Using a less confusing terminology, the next figure shows the inputs to the receptive field of a B/Y retinal ganglion cell in terms of the more neutral S, M, and L terminology for the receptor types.
The retinal ganglion cells that form the B/Y and G/R channels come with a wide variety of receptive field shapes, and each region of the receptive field can make different connections with the various combinations of cone types. Trying to detail all of these specific types and how they relate to overall color information processing quickly becomes too complicated to describe adequately in a short summary so I will stop here. However, I think what I have summarized in my previous post about trichromatic color theory and here about opponent process color theory are enough to provide a basic understanding of the main concepts that have been used to build modern-day, more comprehensive, theories of color vision.
Ronald Boothe
Note 1 — This demonstration might not work properly if viewed on small displays and/or displays that do not accurately display colors. Examples of color adaptation aftereffects, including ones much more complicated than the one shown here can be found in many perception textbooks, and also on websites if you have access to a phone or computer display with sufficient color fidelity to reproduce them properly.
Note 2 — Both ends of the curves are shown truncated in this hypothetical diagram but would actually extend to the limits of the visible spectrum.
Note 3 — “Necessarily” is too strong here, but informal reasoning would be something along the lines: A ratio can be computed at every point along the visible spectrum by dividing the height of A by the height of B. Since white light is a mixture of all wavelengths in the visible spectrum, the effect of white light on A and on B can be obtained by integrating. The overall effect of white light on the A/B opponent channel will be the ratio of the two integrals. This ratio will be equivalent to the mean of the numerator values for the individual wavelength ratios (across all wavelengths in the visible spectrum) dived by the mean of the denominator values for the individual wavelength ratios (across all wavelengths in the visible spectrum). Since the mean has to fall between the extremes, the ratio between the integrals can not be larger than the ratio for any single wavelength or smaller than the smallest. And since the spectral sensitivity functions are continuous there must exist some wavelength that has the same ratio as the ratio of the integrals.
The hypothetical data figures used in this post are reproduced from:
Ronald G. Boothe, Perception of the Visual Environment, Chapter 7 “How are Objective Wavelengths of Light Transformed into Secondary Qualities of Percepts Called Colors?”, Springer-Verlag, New York, 2002.
