Color Perception

Abstract

Color perception, a form of visual perception, is the process by which the visual system assigns hue, saturation, and brightness to surfaces and lights, starting from the rate at which three classes of cone photoreceptor absorb light. Because a single cone confounds the wavelength and the intensity of light in one number, color can be recovered only by comparing cones of different spectral sensitivity, so it is a computation rather than a property read from the world. The article follows this computation from the three cone photopigments and their random retinal mosaic, through the opponent recoding that places every color on a red-green and a blue-yellow axis, to cortical processing and the achievement of color constancy under changing light. Three interactive demonstrations model the trichromatic code, opponent channels, and the discounting of the illuminant.

Keywords: color perception, trichromacy, opponent process, color constancy, cone photopigments

Color perception is the recovery of a stable property of objects, their color, from an unstable and impoverished signal, the light that happens to reach the eye. There is no color in the world outside the observer, only light of different wavelengths; color is the brain's construction, a code it builds by comparing the outputs of a few broadband receptors and then transforming those comparisons into the familiar dimensions of experience. This article follows that construction in order, from the three cone types that first sample the spectrum, through the visual pathway that carries their signals into the brain and the opponent recoding that structures them, to the cortical areas that stabilize color against changing light and the genetic variation that makes one person's color world differ from another's.

What Color Perception Is

Color perception is the set of processes that assign to each region of the visual field a color, described by the three dimensions of hue, saturation, and brightness, from the spectrum of light that region reflects to the eye. Its foundational fact is negative: light itself is not colored. A beam of 700-nanometre light is not intrinsically red; it merely tends to produce the experience of red in a human observer, and the same experience can be produced by a mixture of wavelengths containing no 700-nanometre light at all. Color is therefore a property the visual system attributes to surfaces, a construction rather than a measurement, and two physically different spectra that happen to excite the three cone classes identically are seen as exactly the same color, a phenomenon called metamerism that lies at the root of every color display and print process.

The reason color must be constructed rather than read off is the principle of univariance: a single photoreceptor's response depends on both the wavelength and the intensity of the light it absorbs, collapsed into one number, so a dim light at the cone's best wavelength is indistinguishable from a bright light at a worse one. One receptor class alone is therefore color-blind, able to report only a single quantity of light. Color vision begins the moment the outputs of two or more receptor classes with different spectral sensitivities are compared, because only the ratio of their responses is independent of intensity and specific to the spectral composition of the light. Color is one of the elementary feature maps that vision computes in parallel across the field, which is why a uniquely colored target pops out of a field of distractors at once.

Trichromacy and the Cone Photopigments

Normal human color vision is trichromatic, meaning that three numbers suffice to specify any color and that color matching requires exactly three primaries. The idea long predates its physiological proof: Young proposed that the eye need not hold a separate receptor for every wavelength but only a few color-sensitive elements, and Helmholtz developed this into the theory that three receptor types with overlapping sensitivities span the spectrum (Young, 1802). Trichromacy is thus a fact about the receptors, not about the light, and it sets the dimensionality of the entire color system downstream.

The three receptors are the cones, and their photopigments were eventually measured directly. Microspectrophotometry of single human cones and retinal densitometry established three classes with peak sensitivities in the short, medium, and long wavelengths, confirming the trichromatic hypothesis at the level of the pigment molecule (Wald, 1964; Bowmaker & Dartnall, 1980). Each pigment is an opsin protein bound to a retinal chromophore, and the wavelength of peak absorption is tuned by the opsin's amino-acid sequence. The genes encoding the three opsins were later cloned, revealing that the long- and medium-wavelength genes lie in a tandem array on the X chromosome while the short-wavelength gene sits on chromosome 7, an arrangement whose consequences for individual variation the article returns to below (Nathans et al., 1986a). Table 1 summarizes the three cone classes. The first demonstration makes the trichromatic code concrete, letting the reader sweep a wavelength and watch the three cone excitations, and their ratio, shift together.

The Cone Mosaic

The three cone classes are not laid down in a neat repeating pattern but are interleaved in a mosaic whose arrangement was long inaccessible, because the cones are packed too finely to resolve through the optics of the living eye. Adaptive optics, which corrects the eye's own aberrations, finally allowed the three cone classes to be identified one by one in the living human retina, revealing that the long- and medium-wavelength cones are arranged in patches that appear essentially random, with large differences between individuals in their relative numbers (Roorda & Williams, 1999). The short-wavelength cones are sparse and more regularly spaced, and are absent from the very center of the fovea, which is why the smallest spots are hard to see in blue.

The most striking feature of the mosaic is how much the ratio of long to medium cones varies from one person to the next, ranging from roughly equal numbers to a fourfold or greater excess of long cones, with no corresponding difference in color perception (Hofer et al., 2005). This is a deep clue about how color is computed: if the percept of, say, unique yellow were fixed by the raw numbers of cones, it would vary wildly across people, but it does not, so the visual system must normalize its color code against the long-term average of the input rather than reading cone ratios directly. The stability of color experience across radically different retinas shows that color is built by comparison and adaptation, not by counting photons in fixed proportions.

The Opponent-Process Theory

Trichromacy describes the input but not the code the brain actually uses, and the missing piece is opponency. Introspection had long suggested that the elementary color sensations are organized as antagonistic pairs: Hering noted that no color is ever seen as reddish-green or bluish-yellow, that yellow feels as elementary as red, green, and blue rather than like a mixture, and that colored afterimages take the complementary hue, and he proposed that color is coded in three opponent channels, red versus green, blue versus yellow, and black versus white. The opponent theory and the trichromatic theory were long treated as rivals, but they describe different stages: three cones at the receptors, feeding two chromatic and one achromatic opponent channel just after.

The reconciliation became quantitative through hue cancellation, in which the amount of one color needed to cancel its opponent is measured, yielding the chromatic-response functions that put the red-green and blue-yellow channels on a firm psychophysical footing (Hurvich & Jameson, 1957). Physiology then found the channels in the visual pathway. Recording from cells in the lateral geniculate nucleus, the thalamic relay of the visual system, revealed neurons that are excited by some wavelengths and inhibited by others, spectrally opponent cells that subtract one cone signal from another exactly as the theory requires (De Valois et al., 1966). The blue-yellow channel was traced to a distinct retinal circuit, a bistratified ganglion cell that draws excitation from the short-wavelength cones and opposes it against the long- and medium-wavelength cones (Dacey & Lee, 1994). Figure 1 shows the wiring, and the second demonstration lets the reader drive the three cone signals and watch them recombine into the opponent channels.

Cortical Color Processing

Beyond the opponent channels of the retina and thalamus, color is elaborated in the cortex, beginning in primary visual cortex, area V1. Here the simple center-surround opponency of earlier stages is refined into double-opponent cells, neurons whose receptive fields are opponent for color both at the center and in the surround, so that they respond best to a color boundary, for instance a red region against a green one, rather than to a uniform field (Conway, 2001). Double-opponent cells make color a spatial signal, well suited to marking the edges of objects and to comparing one region against its neighbors, which is the first step toward constancy.

Whether color is then handed to a single dedicated area has been debated since the report of neurons in the prestriate cortex, in and around area V4, that respond to the perceived color of a surface rather than to the raw wavelength reaching the eye, tracking the color an observer would report under changing illumination (Zeki, 1983). Later work has qualified the idea of one color center, showing that color signals are distributed across many visual areas and are intertwined with the coding of form and depth rather than segregated into a pure color stream (Gegenfurtner, 2003). The modern synthesis treats color as computed by machinery that spans the retina, thalamus, and several cortical areas, each transforming the signal, with no single locus where color is finished (Solomon & Lennie, 2007; Gegenfurtner & Kiper, 2003).

Color Constancy

The deepest problem color vision solves is constancy: a surface keeps its apparent color across enormous changes in the light that falls on it, even though the light reflected to the eye, the product of the surface's reflectance and the illuminant, changes just as much. A sheet of white paper reflects far more long-wavelength light under a tungsten bulb than under a blue sky, yet it looks white in both, because the visual system estimates the illuminant and discounts it, recovering a stable descriptor of the surface itself rather than reporting the raw light.

The influential computational account of this achievement is the retinex theory, which proposed that lightness and color are computed separately in long-, medium-, and short-wavelength channels by comparing each region against the spatial average of the whole scene, so that the common effect of the illuminant cancels and only relative reflectances survive (Land & McCann, 1971). A simple and biologically plausible version of this idea is von Kries scaling, in which each cone's gain is adjusted in inverse proportion to the average light it receives, so that a reddish illuminant, which raises the long-wavelength cone's average input, is compensated by turning that cone's gain down. Constancy is never perfect, and the residual failures reveal the estimate at work, but it is good enough that color remains a reliable cue to the identity of objects. The third demonstration puts a fixed surface under a changing illuminant and lets the reader raise the discounting strength to watch the corrected color settle back onto the true reflectance.

Genetics and Individual Variation

Because the long- and medium-wavelength opsin genes sit side by side in a tandem array on the X chromosome and are more than ninety-five percent identical in sequence, the array is unusually prone to misalignment during recombination, which shuffles and deletes genes and creates hybrid opsins with shifted sensitivities. This molecular instability is the direct cause of the common inherited color-vision deficiencies: loss or alteration of the long- or medium-wavelength gene produces the red-green deficiencies that affect roughly eight percent of men, who inherit a single X chromosome, and far fewer women (Nathans et al., 1986b). A person missing one of the two genes is a dichromat, whose color space has two dimensions rather than three; a person with a hybrid gene is an anomalous trichromat, still using three channels but with two of them abnormally close together.

Variation does not stop at deficiency. The same array houses common polymorphisms that shift the long-wavelength pigment's peak by a few nanometres between individuals with normal color vision, and the ratio of long to medium cones in the mosaic varies severalfold, as noted above (Hofer et al., 2005). The visual system absorbs most of this variation by normalizing its opponent code, so that people with quite different retinas still agree closely on where unique hues lie, but at the margins the differences are measurable, and they make color vision one of the clearest cases in which a perceptual experience is shaped by a specific, variable set of genes.

The Functions of Color Vision

Color earns its metabolic cost by making certain visual tasks dramatically easier. Its oldest proposed function is the detection of food against foliage: a red or yellow fruit is nearly invisible to a purely luminance-based system when it matches the brightness of surrounding leaves, but it leaps out to a red-green opponent channel, and the wavelengths at which primate long- and medium-wavelength cones differ are well placed to separate ripe fruit and young leaves from mature foliage. Color also aids the segmentation of scenes more generally, because a region of common color is a cue that its parts belong to one surface, so color helps parse a cluttered image into objects.

Beyond object detection, color supports recognition and memory, because many objects have diagnostic colors that speed their identification, and it carries social and physiological signals, as in the reddening of skin with blood oxygenation and emotion, to which primate color vision appears well tuned (Gegenfurtner & Kiper, 2003). These functions explain why color vision has evolved repeatedly and why its loss, though far from disabling, measurably slows the search for colored targets and the recognition of colored things. Color is not a decorative addition to vision but a channel of information that the visual system exploits at every level from the receptor to the scene.

What the Framework Does and Does Not Explain

The two-stage picture, trichromatic receptors feeding opponent channels, is among the best-established frameworks in perception, but it leaves real questions open. One concerns the number and identity of the opponent axes: the physiological cone-opponent channels measured in the retina and thalamus, long-minus-medium and short-minus-luminance, do not line up with the perceptual opponent axes of unique red, green, blue, and yellow that hue cancellation reveals, so a further cortical transformation must rotate the early channels into the axes of experience, and its mechanism is not settled (Gegenfurtner, 2003). The relation between the physiology and the phenomenology of color is thus still partly open.

A second question concerns constancy and the cortical locus of color. How the visual system estimates the illuminant, whether by simple scene averaging, by finding the brightest region, or by richer inferences about surfaces and light, remains debated, and no single algorithm accounts for human performance across scenes (Land & McCann, 1971). The idea of a dedicated color center in V4 has likewise given way to a more distributed view in which color is inseparable from form and is computed across many areas (Solomon & Lennie, 2007). What is not in dispute is the core architecture: three cone classes set the dimensionality, opponent recoding structures the code, and cortical processing stabilizes it against the light. The debates are about the transformations between these stages, not about the stages themselves.

Worked Example

Consider the three demonstrations with their default constants. In the trichromatic demonstration, the cone excitations are modeled as Gaussian functions of wavelength peaked near 566, 541, and 441 nanometres for the long-, medium-, and short-wavelength cones, each with a spread, so the excitation of a cone is the exponential of minus one half times the squared difference between the stimulus wavelength and the cone's peak, divided by the squared spread. For a monochromatic light at 580 nanometres the long cone gives 0.95, the medium cone gives 0.69, and the short cone gives essentially zero, which normalized to a sum of one is a ratio of long to medium to short of 0.58 to 0.42 to 0.00. That the long and medium cones are both strongly and nearly equally driven, with the short cone silent, is exactly the receptor condition the next stage reads as yellow.

The opponent demonstration takes three cone excitations and forms the channels directly: luminance is the sum of the long and medium signals, the red-green channel is the long minus the medium signal, and the blue-yellow channel is the short signal minus the average of the long and medium signals. With long, medium, and short excitations of 0.8, 0.5, and 0.2, the luminance channel is 1.3, the red-green channel is 0.8 minus 0.5, or plus 0.3, toward red, and the blue-yellow channel is 0.2 minus 0.65, or minus 0.45, toward yellow, giving an orange appearance. The color-constancy demonstration then fixes a surface whose reflectances in the three bands are 0.60, 0.40, and 0.30 and places it under a warm illuminant with components 1.30, 1.00, and 0.70. The light the cones catch is the product of reflectance and illuminant, 0.78, 0.40, and 0.21, which drifts toward the long wavelengths; dividing each channel by the illuminant, full von Kries discounting, recovers exactly 0.60, 0.40, and 0.30, the true reflectance, while half-strength discounting recovers 0.68, 0.40, and 0.25, a partial correction that still carries a residual warm cast.

Discussion

Color perception is best understood as a sequence of transformations that turn an intractable physical signal into a stable and useful code. The light reaching the eye confounds surface and illuminant and offers, at each point, only the absorption rates of three broadband cones, each of which by the principle of univariance cannot on its own distinguish wavelength from intensity (Wald, 1964). Trichromacy fixes the dimensionality of the problem at three; opponent recoding restructures those three signals into an achromatic and two chromatic channels whose differences are what the brain actually reads, explaining the forbidden colors and the complementary afterimages that first suggested the scheme (Hurvich & Jameson, 1957; De Valois et al., 1966).

The cortical stages then convert this local code into a representation of surfaces, using double-opponent cells to make color a spatial, edge-marking signal and distributed cortical processing to discount the illuminant and achieve constancy (Conway, 2001; Land & McCann, 1971). That the whole system rests on a handful of variable opsin genes, and yet delivers a color world stable enough that people with fourfold differences in cone ratios agree on where yellow lies, is the strongest evidence that color is computed by comparison and adaptation rather than read from the receptors (Roorda & Williams, 1999; Hofer et al., 2005). The open problems, how the early cone-opponent axes become the perceptual hue axes and how the illuminant is estimated, concern the transformations between well-established stages, and the enduring lesson is that the immediacy of color conceals one of the deepest constructive computations in vision.

Glossary

Anomalous trichromacy.

A color-vision variant in which all three cone classes are present but two have abnormally similar sensitivities, compressing one opponent axis and weakening color discrimination.

Chromatic adaptation.

The adjustment of the visual system's sensitivity to the prevailing color of illumination, so that the response to a surface is referred to the average light rather than the raw light.

Color constancy.

The tendency of a surface to keep its apparent color across changes in the illuminant, achieved by estimating and discounting the light that falls on it.

Cone mosaic.

The spatial arrangement of the short-, medium-, and long-wavelength cones across the retina, which is largely random and varies greatly between individuals.

Cone.

A photoreceptor of the retina that operates in daylight and comes in three spectral classes, whose comparison provides the basis of human color vision.

Dichromacy.

A form of color vision using only two cone classes rather than three, reducing color space to two dimensions, as in common red-green color-vision deficiencies.

Double-opponent cell.

A cortical neuron whose receptive field is color-opponent in both center and surround, responding best to a chromatic edge and forming the basis of spatial color processing.

Hue cancellation.

A method that measures an opponent channel by finding how much of an opposing color must be added to null a given hue, yielding the chromatic-response functions.

Metamerism.

The phenomenon by which two physically different light spectra that excite the three cone classes identically are seen as exactly the same color.

Opponent process.

The coding of color in antagonistic channels, red versus green and blue versus yellow, plus a black-white channel, so that the paired colors cannot be seen at once.

Opsin.

The protein component of a photopigment, whose amino-acid sequence tunes the wavelength of peak light absorption and so defines a cone's spectral class.

Photopigment.

The light-absorbing molecule in a photoreceptor, an opsin bound to a retinal chromophore, whose absorption spectrum sets the receptor's sensitivity to wavelength.

Retinex.

A computational theory of color constancy in which lightness is computed separately in each waveband by comparing regions across the scene, cancelling the illuminant.

Spectral sensitivity.

The profile describing how strongly a receptor responds to each wavelength of light, broad and overlapping across the three cone classes.

Trichromacy.

The property of normal human color vision that three cone classes, and hence three numbers, suffice to specify any color and to match any light.

Univariance.

The principle that a single photoreceptor's output confounds the wavelength and the intensity of light, so one receptor class alone cannot signal color.

Von Kries adaptation.

A model of color constancy in which each cone class independently scales its gain in inverse proportion to the average light it receives, discounting the illuminant.

Key Researchers

Thomas Young. English polymath of the early nineteenth century; in his 1802 Bakerian Lecture he argued that the eye need not hold a receptor for every wavelength but only a few color-sensitive elements, first stating the trichromatic principle later formalized as the Young-Helmholtz theory. Wikipedia) - Wikidata

Hermann von Helmholtz. German physicist and physiologist; he developed Young's conjecture into a quantitative trichromatic theory in which three overlapping receptor sensitivities span the spectrum, setting out the framework in his Handbuch der physiologischen Optik. Wikipedia - Wikidata

Ewald Hering. German physiologist; he proposed the opponent-process theory of color, holding that color is coded in antagonistic red-green, blue-yellow, and black-white channels, to explain the forbidden colors and complementary afterimages that trichromacy could not. Wikipedia - Wikidata

Leo M. Hurvich. Color-vision psychologist at the University of Pennsylvania; with Dorothea Jameson he revived and quantified Hering's opponent theory through hue-cancellation experiments, measuring the chromatic-response functions of the opponent channels. Wikipedia - Wikidata

Dorothea Jameson. Vision scientist and National Academy of Sciences member at the University of Pennsylvania; with Hurvich she derived the quantitative opponent-process account of color, modeling how the opponent channels underlie color appearance and adaptation. Wikipedia - Wikidata

Jeremy Nathans. Molecular geneticist at the Johns Hopkins University School of Medicine; he cloned the human cone-opsin genes, showed that the long- and medium-wavelength genes lie in a variable array on the X chromosome, and traced red-green color deficiency to that array. ORCID - Faculty Page - Wikipedia

Semir Zeki. Neurobiologist at University College London; he identified neurons in and around area V4 that respond to the perceived color of a surface rather than to the raw wavelength reaching the eye, linking cortical processing to color constancy. ORCID - Faculty Page - Wikipedia

Karl R. Gegenfurtner. Vision scientist at Justus Liebig University Giessen; he synthesized the cortical machinery of color vision, showing that color signals are distributed across many visual areas and analyzing how color supports object recognition and scene segmentation. ORCID - Faculty Page - Google Scholar

Frequently Asked Questions

What is color perception?
It is the process by which the visual system assigns hue, saturation, and brightness to surfaces and lights, constructing color as a neural code by comparing the responses of three cone classes rather than reading it directly from the light (Wald, 1964).

Is color a property of light?
No; light has wavelength but not color, and color is the brain's construction, which is why two physically different spectra that excite the three cone classes identically, called metamers, are seen as exactly the same color.

Why are three cone types needed for color vision?
Because a single cone obeys the principle of univariance, confounding wavelength and intensity in one number, so color begins only when the outputs of cones with different spectral sensitivities are compared, and three classes make normal human vision trichromatic (Young, 1802).

What is the opponent-process theory?
It holds that cone signals are recoded into antagonistic channels, red versus green and blue versus yellow plus a black-white channel, which explains why no color looks reddish-green or bluish-yellow and why afterimages take the complementary hue (Hurvich and Jameson, 1957).

How do trichromatic and opponent theories fit together?
They describe successive stages rather than rivals: three cone classes sample the spectrum at the receptors, and their signals are then subtracted into opponent channels in the retina and thalamus, a scheme confirmed by spectrally opponent neurons (De Valois et al., 1966).

What is color constancy?
It is the tendency of a surface to keep its apparent color under different lighting, achieved when the visual system estimates the illuminant and discounts it, as captured by the retinex theory and von Kries gain scaling (Land and McCann, 1971).

What causes color blindness?
The long- and medium-wavelength opsin genes sit in a nearly identical array on the X chromosome, prone to misalignment that deletes or hybridizes genes, so the common red-green deficiencies affect roughly eight percent of men and far fewer women (Nathans et al., 1986b).

Where in the brain is color processed?
Color is computed across the retina, the thalamus, and several cortical areas from V1 to V4, with double-opponent cells marking chromatic edges and distributed cortical processing achieving constancy, rather than in a single color center (Zeki, 1983; Solomon and Lennie, 2007).

References