Color Vision

Abstract

Color vision is a form of ocular vision: the capacity to distinguish light by its wavelength composition and so to see the world in color rather than in shades of grey. The vision sciences study it because color is a constructed property — the brain assigns it from the ratios of activity in only three cone types, not from any wavelength label carried by light itself. This article traces the two-stage architecture that makes color possible: the trichromatic front end, in which three cone photopigments sample the spectrum, and the opponent recoding that differences their signals into red–green, blue–yellow, and light–dark channels. It covers the photopigment genetics behind normal vision and color blindness, the living cone mosaic, color constancy, cortical processing, and individual differences. Three interactive demonstrations let the reader mix colored lights, generate an afterimage, and simulate color blindness.

Keywords: color vision, trichromacy, opponent process, color perception, ocular vision

Color vision is the ability to discriminate lights on the basis of their spectral composition independently of their intensity, and to organise the resulting sensations into the familiar dimensions of hue, saturation, and brightness (Gegenfurtner & Kiper, 2003). Its central puzzle is that the eye has no wavelength detector. Each cone obeys the principle of univariance: it counts photons and reports a single number, so a change in wavelength and a change in intensity are confounded within any one receptor and cannot be told apart (Stockman & Sharpe, 2000). Color therefore cannot be read from a single receptor; it must be computed by comparing the outputs of receptors with different spectral sensitivities. This is why human color vision rests on exactly three cone classes, and why the whole visible gamut can be matched by mixing just three primaries — the empirical fact from which Thomas Young inferred a three-receptor theory two centuries ago (Young, 1802). What makes color a subject for cognitive psychology rather than only optics is that the percept is inferred: the same physical spectrum can look like different colors under different surrounds and illuminants, and physically different spectra can look identical, so color is a property the brain assigns rather than one it merely measures (Shevell & Kingdom, 2008). This article should be read alongside color perception, which treats the subjective experience and naming of color; the present article concerns the sensory mechanism that makes those experiences possible.

What Color Vision Is

Color vision is the discrimination and organisation of light by wavelength. It is best defined operationally: an observer has color vision if they can tell apart two lights that differ only in spectral composition, matched for intensity (Stockman & Sharpe, 2000). This operational test matters because color is not a physical property of light — light has a spectral power distribution, but “redness” is a response of the visual system, assigned to certain distributions by the machinery described below (Conway, 2009). The dimensionality of that response is the first great fact about color: normal human vision is three-dimensional, so any color can be specified by three numbers and matched by three suitably chosen primaries.

It is worth separating color vision from color perception. Color vision refers to the sensory apparatus — the receptors and neural channels that make wavelength discrimination possible — while color perception refers to the resulting experience, including how colors are categorised, named, and judged in context. The two are tightly coupled but dissociable: two people with identical cone complements can carve the spectrum into different named categories, and the same eye can yield different perceived colors depending on the surround (Witzel & Gegenfurtner, 2018). Color vision also serves biological functions beyond aesthetics: for primates it aids the detection of ripe fruit and young leaves against foliage and supports the reading of social signals such as blushing and blanching of the skin (Mollon, 1989). Across mammals, color vision varies widely, and the primate three-cone system is an evolutionary specialisation rather than the mammalian norm (Jacobs, 2009).

The Trichromatic Foundation

The first stage of color vision is trichromacy: the presence of three cone photopigments with different spectral sensitivities. Young proposed on purely behavioural grounds that the eye must contain a small number of receptor types, each tuned to a different part of the spectrum, because a small set of primary lights could reproduce every visible color (Young, 1802). Hermann von Helmholtz developed the idea quantitatively, and the resulting Young–Helmholtz theory held that three receptor classes, broadly tuned and overlapping, encode wavelength by the pattern of their relative activity. The theory was vindicated directly when microspectrophotometry measured the absorption spectra of single human cones and found exactly three classes, with peak sensitivities in the short, medium, and long wavelengths (Bowmaker & Dartnall, 1980; Dartnall et al., 1983). Precise psychophysical estimates of the M- and L-cone spectral sensitivities, tied to observers of known genotype, later refined these curves (Stockman & Sharpe, 2000).

Trichromacy has a profound consequence: metamerism. Because each cone reduces an entire spectrum to a single number, a light is represented in the eye by just three values — the excitations of the S, M, and L cones. Any two spectra that produce the same three excitations are indistinguishable, however different their physical composition. These matching pairs, called metamers, are why a color monitor with only three primaries can reproduce the appearance of a continuous natural spectrum, and why color mixing is additive and lawful (Gegenfurtner & Kiper, 2003). Trichromacy also fixes the dimensionality of the color space at three, which is the deep reason color has exactly three perceptual dimensions rather than the near-infinite dimensionality of the physical spectrum. The spectral sensitivity of each cone — its probability of catching a photon as a function of wavelength — is therefore the foundational data of the whole system.

Opponent Processing

Trichromacy describes the receptors, but it does not describe experience. Color does not feel three-dimensional in the way three cones might suggest: there is no such thing as a reddish green or a bluish yellow, and the hues seem organised into opposed pairs. Four hues — red, green, blue, and yellow — look phenomenally pure, or unique, in that they appear to contain no trace of any other hue, while every other color is seen as a mixture of these. Ewald Hering argued from these phenomenological facts that color is coded in opponent channels, with red opposed to green and blue to yellow. The two views — trichromatic receptors versus opponent experience — were reconciled when Leo Hurvich and Dorothea Jameson quantified the opponent signals using hue cancellation, measuring how much of one hue is needed to null its opposite, and showed that the data demand an opponent stage built on top of the three cones (Hurvich & Jameson, 1957). Physiology then found the opponent cells themselves: neurons in the lateral geniculate nucleus that are excited by some wavelengths and inhibited by others, differencing cone signals rather than simply summing them (De Valois et al., 1966).

The modern account is therefore two-stage. Three cone types transduce light (the trichromatic stage), and their outputs are then recoded by differencing into three post-receptoral channels: a red–green channel (roughly L − M), a blue–yellow channel (roughly S − (L + M)), and an achromatic light–dark channel (roughly L + M) (Solomon & Lennie, 2007). This recoding is efficient — it decorrelates the heavily overlapping cone signals — and it explains the opponent process structure of experience, including negative afterimages, in which fixating a color fatigues one limb of an opponent channel so that a neutral field then appears in the complementary hue. The neat correspondence between the cardinal cone-opponent axes and perceived unique hues is imperfect, however, and the classical identification of the physiological axes with red–green and blue–yellow appearance has been challenged by recent cross-cultural and computational work (Conway et al., 2023).

The Living Cone Mosaic

For most of the twentieth century the arrangement of the three cone types in the retina could only be inferred. Adaptive optics changed that: by correcting the eye's own optical aberrations, David Williams and colleagues imaged the cone mosaic in the living human eye and, using selective bleaching, classified each cone as S, M, or L (Roorda & Williams, 1999). The images revealed a striking irregularity. The S cones form a sparse, semi-regular lattice making up under a tenth of the total, while the L and M cones are interspersed in patches whose proportions vary enormously between people (Hofer et al., 2005).

The most surprising finding was the range of L:M cone ratios across individuals with entirely normal color vision. Some retinas contain roughly twice as many L as M cones; others are the reverse; ratios differing several-fold are all compatible with normal color naming and discrimination (Hofer et al., 2005). This is a deep point about the system: the perceptual red–green balance is remarkably stable despite large differences in the raw receptor sampling, which means the opponent stage must be calibrated by experience with the world rather than wired to a fixed anatomy. The living-mosaic work thus tied the psychophysics of trichromacy to a concrete, countable retinal substrate and exposed how much normalisation the brain performs.

Photopigment Genetics and Color Blindness

The molecular basis of trichromacy is now known in detail. The genes encoding the cone opsins were identified in the 1980s: the S-cone gene sits on chromosome 7, while the M- and L-cone genes lie in a head-to-tail tandem array on the X chromosome (Nathans et al., 1986). This arrangement — two highly similar genes adjacent on a single chromosome — makes the array prone to unequal recombination, which shuffles, deletes, and creates hybrid genes. The result is that variation and defect in red–green color vision are common and sex-linked (Neitz & Neitz, 2011).

The consequences follow directly. A dichromacy arises when one of the three photopigment classes is missing: protanopia (no L), deuteranopia (no M), or the rare tritanopia (no S). More common are the anomalous trichromacies, in which a hybrid pigment shifts the M or L sensitivity so that the two long-wavelength cones become more alike, compressing the red–green dimension. Because the M and L genes are X-linked, these red–green deficiencies affect roughly 8% of men but under 1% of women (Neitz & Neitz, 2011). The same understanding of the pigment genes made a remarkable therapy possible: introducing a missing opsin gene into the cones of dichromatic adult monkeys restored trichromatic behaviour, showing that the adult visual system can extract a new color dimension from a receptor change without rewiring in infancy (Mancuso et al., 2009). The circuitry that supports conscious color and how it accommodates such changes remains an active question (Neitz & Neitz, 2017).

The principal inherited deficiencies, and the pigment changes that cause them, are summarised in Table 1.

Color Constancy

A surface reflects a fixed proportion of the light striking it, but the light itself changes constantly — bluish in shade, reddish at sunset, greenish under foliage — so the spectrum reaching the eye from any object shifts throughout the day. Yet objects keep a roughly stable color: a banana looks yellow at noon and at dusk. This achievement is color constancy, the discounting of the illuminant to recover the surface's own reflectance (Foster, 2011). Constancy is never perfect, but it is good enough that we treat color as a property of things rather than of the light, which is precisely what makes color useful for identifying objects.

Edwin Land's retinex theory offered an influential computational account: the visual system estimates the illuminant from spatial comparisons across the whole scene — in effect normalising each region against a running average — and assigns color from these ratios rather than from absolute cone excitations (Land, 1977). Modern work treats constancy as an inference problem in which the brain combines several cues — the space-average color of the scene, the brightest regions, specular highlights, and the statistics of natural illuminants — to estimate and remove the illuminant (Foster, 2011). Constancy is inseparable from the way color behaves in complex scenes, where a patch's appearance depends heavily on its surround, and the same local spectrum can look strikingly different in different contexts (Shevell & Kingdom, 2008). Adaptation is a key mechanism here: the visual system continually reweights its sensitivity to the prevailing color, both maintaining constancy and producing the aftereffects that reveal the opponent channels at work (Webster, 2015).

Cortical Color Processing

Beyond the retina and lateral geniculate nucleus, color is elaborated in the visual cortex. The cone-opponent signals arriving at primary visual cortex (V1) are transformed there and in downstream areas into representations that align more closely with perceived color than the cardinal cone-opponent axes do (Solomon & Lennie, 2007). Specialised color-responsive regions have been identified in the ventral visual pathway, and their activity tracks perceived rather than merely physical color, linking the machinery of color vision to the higher processes of recognition and categorisation (Conway, 2009). This cortical stage is where color vision meets cognition: the transformation from a decorrelated, efficient sensory code to a representation organised around behaviourally meaningful categories such as the basic color terms (Gegenfurtner & Kiper, 2003). How far the boundaries of those categories are fixed by the biology of the visual system, and how far they are shaped by language and culture, is a question the cortical work has reopened (Skelton et al., 2017).

Worked Example

Because each cone reports a single number, a light is represented in the eye by just three values — the excitations of the S, M, and L cones — and the opponent stage forms three differences of them. Working an example through the transform shows both why metamers are inevitable and how a percept can shift.

Suppose a light produces normalised cone excitations of L = 0.60, M = 0.50, and S = 0.20 (each on a 0–1 scale). The three post-receptoral channels compute, in this simplified model: an achromatic light–dark signal of L + M = 0.60 + 0.50 = 1.10; a red–green signal of L − M = 0.60 − 0.50 = +0.10, slightly toward red; and a blue–yellow signal of S − (L + M) / 2 = 0.20 − 0.55 = −0.35, toward yellow. The percept is a bright, slightly reddish yellow.

Now consider metamerism. A physically different light — a different spectral power distribution entirely — that happens to excite the cones to the same (0.60, 0.50, 0.20) will yield identical channel outputs of (1.10, +0.10, −0.35) and so will be indistinguishable from the first. Nothing downstream of the cones can recover the spectral difference, because the three cone numbers are all the information that survives transduction. This is the arithmetic root of metamerism and of why three primaries suffice.

Finally, hold L fixed and raise M from 0.50 to 0.60, as a shift in relative cone stimulation might. The red–green signal becomes L − M = 0.60 − 0.60 = 0, so the light loses its reddish tinge and sits at the neutral point of the red–green axis, while the blue–yellow signal moves to 0.20 − 0.60 = −0.40, slightly more yellow. A change of only 0.10 in one cone's excitation has abolished the red–green content of the color. The lesson is quantitative: color lives in the differences between cone signals, so small changes in the balance of cone excitation, not their absolute level, drive changes in hue — which is exactly why the opponent recoding is the operative stage for color appearance.

Discussion

Color vision earns its place in cognitive psychology because it is a worked example of perception as inference. The receptors deliver an impoverished, ambiguous signal — three numbers per location, each confounding wavelength with intensity — and everything we experience as color is computed from comparisons among those numbers and across the scene (Gegenfurtner & Kiper, 2003). The two-stage architecture, trichromatic then opponent, is a case study in efficient coding: the second stage discards the redundancy in the heavily overlapping cone signals, and the resulting channels map onto the structure of experience (Solomon & Lennie, 2007). Color constancy extends the same logic, treating the proximal spectrum as evidence about a distal surface under an unknown illuminant and inferring the surface color (Foster, 2011).

The account has open seams. The correspondence between the physiological cone-opponent axes and the phenomenological unique hues is looser than the textbook story implies, and the very identification of red–green and blue–yellow as the fundamental axes has been questioned (Conway et al., 2023). The stability of color perception across large individual differences in cone ratios shows that experience calibrates the system, but how that calibration works is not settled (Hofer et al., 2005). And the boundary between color vision and color cognition — where sensory coding ends and categorisation, memory, and language begin — remains contested, nowhere more than in the debate over how far color categories are universal (Skelton et al., 2017; Witzel & Gegenfurtner, 2018).

Current Directions

Contemporary research is moving on three fronts. The first is individual differences: once treated as measurement noise around a normal observer, variation in color perception is now studied in its own right, from anomalous trichromacy to the substantial differences among color-normal observers in unique hues, discrimination, and categorisation (Emery & Webster, 2019; Bosten, 2022). The scale of this diversity has forced a reconsideration of what a “normal” color observer even is (Bosten, 2022). The second is the origin of color categories: infant and cross-cultural studies suggest that the broad divisions of color space have a biological basis in the visual system, even as language sharpens and shifts their boundaries, reframing the old universalism-versus-relativism debate in developmental terms (Skelton et al., 2017). The third is a re-examination of the opponent framework itself: large cross-linguistic datasets and analyses of color appearance have challenged Hering's opponent-colors theory as the correct description of color experience, suggesting the mapping from cone-opponent physiology to perceived hue is more complex than the classical model allows (Conway et al., 2023; Witzel & Gegenfurtner, 2018). Running through all three is a methodological shift toward measuring color under natural conditions and across diverse populations rather than in the reduced laboratory displays that defined the field for a century.

Common Misconceptions

Color is a physical property of light.

Light has a spectral power distribution; color is the visual system's response to it. Physically different spectra can look identical (metamers) and the same spectrum can look different in different contexts, so color is assigned by the brain, not carried by the light (Shevell & Kingdom, 2008; Conway, 2009).

Each cone type detects one color — red, green, or blue.

Cones are broadly tuned and heavily overlapping, and each is univariant, so no cone signals a color by itself. Color comes from the ratio of excitations across cone types, which is why the “red, green, blue” labels for the L, M, and S cones are misleading (Stockman & Sharpe, 2000).

People with the same normal color vision have the same retina.

The ratio of L to M cones varies several-fold across color-normal individuals, yet their color perception is closely matched — evidence that experience calibrates the opponent stage rather than the anatomy fixing it (Hofer et al., 2005).

Glossary

Adaptation.

The continual reweighting of visual sensitivity to the prevailing stimulation; chromatic adaptation reduces sensitivity to the ambient color, aiding constancy and producing aftereffects.

Anomalous trichromacy.

A form of color vision that retains three cone pigments but with a hybrid M or L pigment whose shifted sensitivity compresses the red-green dimension; the most common inherited color-vision deficiency.

Color constancy.

The tendency for a surface to keep a stable perceived color despite changes in the spectrum of the illuminating light; the visual system discounts the illuminant to recover reflectance.

Cone mosaic.

The spatial arrangement of the S, M, and L cones across the retina; imaged in the living eye with adaptive optics, its L:M ratio varies severalfold among people with normal color vision.

Cone.

A photoreceptor of the retina that operates in daylight and contains one of three photopigments; the three cone classes are the basis of trichromatic color vision.

Dichromacy.

A form of color vision with only two functioning cone photopigments instead of three, reducing color to a two-dimensional space; protanopia, deuteranopia, and tritanopia are the three types.

Hue.

The attribute of a color that lets it be named red, green, blue, and so on; together with saturation and brightness it is one of the three perceptual dimensions of color.

Metamerism.

The phenomenon whereby two physically different spectra look identical because they produce the same excitations in the three cones; the direct consequence of trichromacy.

Opponent process.

The recoding of cone signals into differencing channels — red–green, blue–yellow, and light–dark — at a stage after the receptors; it explains opposed hues and complementary afterimages.

Opsin.

The protein component of a cone or rod photopigment; the genes encoding the M- and L-cone opsins lie in a tandem array on the X chromosome, the site of the common red-green deficiencies.

Photopigment.

The light-absorbing molecule in a photoreceptor — an opsin protein bound to a chromophore — whose absorption spectrum sets that receptor's spectral sensitivity.

Retinex.

Land's computational theory of color constancy in which the visual system estimates surface color from spatial comparisons of cone signals across the scene rather than from absolute excitations.

Spectral sensitivity.

The probability that a receptor absorbs a photon as a function of the photon's wavelength; the three cone spectral sensitivities are the input data of color vision.

Trichromacy.

The possession of three cone photopigment classes, making color vision three-dimensional so that three primaries suffice to match any color; the normal human condition.

Unique hues.

The four hues — red, green, blue, and yellow — that appear phenomenally pure, containing no trace of any other hue; the perceptual landmarks of the opponent channels, from which Hering inferred opponent coding.

Univariance.

The principle that a single photoreceptor's output varies along only one dimension, so it cannot by itself distinguish a change in wavelength from a change in intensity.

Key Researchers

Jenny M. Bosten. Vision scientist at the University of Sussex; she studies individual differences in human color perception, color categorization, anomalous trichromacy, and the neural mechanisms of color vision. ORCID - Google Scholar

Bevil R. Conway. Vision scientist at the National Eye Institute (NIH); he investigates the neural basis of color in the visual cortex, identifying color-processing regions and linking chromatic signals to cognition and language. ORCID - Faculty Page - Wikipedia

Karl R. Gegenfurtner. Psychologist at Justus Liebig University Giessen; he studies the cortical mechanisms of color vision, color constancy, and the relation between low-level sensory processing and visual cognition. Faculty Page - Google Scholar

Leo M. Hurvich. Psychologist at the University of Pennsylvania; with Dorothea Jameson he developed the quantitative opponent-process theory of color vision and the hue-cancellation technique. Wikipedia - Wikidata

Dorothea Jameson. Psychologist at the University of Pennsylvania; with Leo Hurvich she developed hue cancellation and the data that established the modern opponent-process theory of color vision. Wikipedia - Wikidata

John D. Mollon. Visual neuroscientist at the University of Cambridge; he works on human colour vision, cone photopigments, and the evolution of primate trichromacy, and developed the Cambridge Colour Test. Faculty Page - Wikipedia

Jay Neitz. Color vision scientist at the University of Washington; he is known for the photopigment genetics of color vision and gene-therapy approaches to curing red–green color blindness. Faculty Page - Wikipedia

Maureen Neitz. Vision scientist at the University of Washington; she specialises in the photopigment gene genetics underlying color-vision deficiencies and gene-therapy approaches to color blindness. ORCID - Faculty Page

Michael A. Webster. Psychologist at the University of Nevada, Reno; he works on color appearance, visual adaptation, and individual differences in color perception, including adaptation that recalibrates color and blur. Faculty Page - Google Scholar

David R. Williams. Vision scientist at the University of Rochester; he built the first adaptive-optics system for the human eye, enabling high-resolution imaging of the living cone mosaic. Faculty Page - Lab Page

Frequently Asked Questions

What is color vision?
Color vision is the ability to distinguish lights by their wavelength composition, independently of intensity, and to organise the result into hue, saturation, and brightness. It depends on comparing the outputs of cones with different spectral sensitivities, because no single cone can signal color (Gegenfurtner & Kiper, 2003; Stockman & Sharpe, 2000).

Why do humans have three cone types?
Three broadly tuned, overlapping cone photopigments let the visual system encode wavelength by the ratio of their activity, which is enough to match any color with three primaries. This trichromatic design was inferred behaviourally by Young and confirmed by direct measurement of cone spectra (Young, 1802; Bowmaker & Dartnall, 1980).

What is metamerism?
Metamerism is the fact that two physically different light spectra can look identical because they excite the three cones equally. It follows directly from trichromacy and is why three-primary displays can reproduce natural colors (Gegenfurtner & Kiper, 2003).

What is the opponent-process theory?
It holds that cone signals are recoded into opposed channels (red versus green and blue versus yellow, plus light versus dark) at a stage after the receptors. Hurvich and Jameson quantified these channels, reconciling Hering's opponent experience with Young and Helmholtz trichromacy (Hurvich & Jameson, 1957; De Valois et al., 1966).

Why is red-green color blindness so common in men?
The M- and L-cone opsin genes lie in a tandem array on the X chromosome, where they recombine unequally to produce missing or hybrid pigments. Because men have a single X chromosome, these red-green deficiencies affect about 8% of men but under 1% of women (Nathans et al., 1986; Neitz & Neitz, 2011).

What is color constancy?
Color constancy is the tendency for objects to keep a stable perceived color even as the illuminating light changes. The visual system estimates and discounts the illuminant from comparisons across the scene, as in Land's retinex theory (Land, 1977; Foster, 2011).

Do all people with normal color vision see color the same way?
No. The ratio of L to M cones varies several-fold across color-normal people, and unique hues and category boundaries differ substantially, so individual differences in color perception are large and are now studied in their own right (Hofer et al., 2005; Bosten, 2022).

Can color blindness be cured?
In adult monkeys with red-green color blindness, adding the missing opsin gene to the cones restored trichromatic behaviour, showing the adult visual system can use a new color signal. Gene therapy in humans is being pursued but is not yet an established treatment (Mancuso et al., 2009; Neitz & Neitz, 2017).

References