In plain terms, color constancy is why a banana looks yellow at noon, at sunset, and under your kitchen's fluorescent light — even though the light bouncing off it into your eyes is physically different in each case. The light that reaches the eye is a mixture of two things the visual system cannot separately measure: the color of the surface and the color of the light falling on it. Pull a white shirt from bright daylight into a warm lamp-lit room and the light it reflects shifts from bluish to orange, yet you go on seeing a white shirt. Color constancy is the set of processes that achieves this — discounting the illumination to recover the stable color of the object itself. This article explains what color constancy is and how it is measured, why it is a hard computational problem, the mechanisms that solve it from the retina up to the scene, where it breaks down (as in "the dress"), and what current research and open questions look like.
Color constancy is the tendency for the perceived color of a surface to remain approximately stable despite changes in the spectrum and intensity of the light illuminating it (Foster, 2011). It is one of the perceptual constancies — alongside size, shape, and lightness constancy — by which the visual system recovers stable properties of objects from a constantly changing retinal image. Constancy matters because the light reaching the eye, the color signal, is the product of the surface's spectral reflectance and the illuminant's spectral power distribution; to perceive the object's own color, the brain must somehow undo the contribution of the light. This is the central problem of color vision, and it connects work from the wavelength tuning of cones in the retina to the color-selective machinery of cortical area V4. Importantly, constancy is rarely perfect: it is usually partial, and how complete it is depends heavily on the richness of the scene (Witzel & Gegenfurtner, 2018).
What Is Color Constancy?
Sit a bowl of fruit by a window over the course of a day. At noon the light is bluish and bright; at sunset it is warm and dim; under a lamp at night it is oranger still. The wavelength composition of the light each piece of fruit reflects changes enormously across these conditions, yet the apples keep looking red and the lemons yellow. That stability is color constancy, and it is so automatic that its existence is easy to overlook — the effect is precisely that we do not notice the illumination changing the colors of things (Foster, 2011).
Constancy is best understood as a graded achievement rather than an all-or-nothing fact. Researchers quantify it with a constancy index, which runs from 0 (the percept tracks the raw light, no constancy) to 1 (the percept tracks the surface perfectly, complete constancy). Real performance falls in between and varies with conditions: in careful experiments the degree of constancy can be moved anywhere from roughly 10% to over 80% by changing how much scene information is available (Kraft & Brainard, 1999). Constancy is strongest in rich, three-dimensional scenes with many surfaces, highlights, and clear cues to the light, and weakest for an isolated patch viewed in a void (Foster, 2011).
Why Color Constancy Matters
Color constancy is not just a laboratory curiosity; it is the reason color is a useful signal at all. If an object's apparent color swung with every change in lighting, color could not help you find ripe fruit, recognize a familiar face, or pick your car out of a parking lot. Stable surface color is a dependable cue to object identity and material, which is plausibly why the visual system invests so much machinery in computing it (Witzel & Gegenfurtner, 2018).
The same problem reappears wherever images are captured or reproduced. A camera's white balance is an engineered version of color constancy: the device estimates the color of the light and corrects the image so that whites look white — and it draws on the same computational ideas (recover the surface, discount the illuminant) developed in vision science (Maloney & Wandell, 1986; Brainard & Freeman, 1997). Color management for displays and printing rests on related models of how appearance depends on the viewing illuminant (Brainard & Maloney, 2011). And the limits of constancy have everyday consequences for design and retail: a paint chip or a shirt that match under store lighting can look noticeably different in daylight — a failure of color matching across illuminants that has been measured directly in natural scenes (Foster, Amano, Nascimento, & Foster, 2006).
The Problem: Why Constancy Is Hard
The deep difficulty is that color constancy is an inverse problem, and a mathematically underdetermined one. The signal a cone registers from a patch of surface depends on the product of two spectra the visual system never sees in isolation: the spectral power distribution of the light and the spectral reflectance of the surface. Many different combinations of light and surface can yield exactly the same cone responses, so the retinal signal alone does not determine what color the object is — a reddish surface under neutral light and a neutral surface under reddish light can look identical at a single point (Smithson, 2005).
Because the problem is underdetermined, it can be solved only by bringing extra assumptions to bear — by exploiting statistical regularities of natural lights and surfaces, or by reading cues in the scene that betray the illuminant. Every theory of color constancy is, at bottom, a proposal about which assumptions the visual system uses and how it applies them (Smithson, 2005; Maloney & Wandell, 1986).
Discounting the Illuminant
The oldest idea about how the brain solves this problem is also the most durable. Helmholtz proposed in the nineteenth century that perception involves unconscious inference: the visual system makes rapid, automatic, unnoticed judgments about the scene, and in the case of color it effectively estimates the illumination and then "discounts" it, attributing the remaining variation to the surface (Helmholtz, 1867/1924). On this view, seeing a stable color is not a passive readout of the light but an active inference about its most probable cause. Helmholtz's framing — estimate the light, then subtract its influence — still structures essentially every modern account, from receptor-level adaptation to full Bayesian models (Smithson, 2005).
Mechanisms: From Receptors to Scenes
Color constancy is not one mechanism but a layered set of them, operating from the photoreceptors up to high-level knowledge. It is useful to organize them as sensory, computational, and cognitive contributions (Smithson, 2005).
Chromatic adaptation. The earliest and simplest contribution is receptoral. When the illumination shifts toward, say, longer wavelengths, the long-wavelength cones are more strongly stimulated and their gain is turned down, while the other cone classes adjust independently. This independent rescaling of the three cone signals is the classical von Kries coefficient law (Kries, 1905/1970), and it goes a long way toward stabilizing color across changes in the light — it is why a room lit with warm bulbs stops looking orange after a few minutes. Adaptation is powerful but limited: it treats the whole image alike and cannot, by itself, account for the constancy seen in complex scenes (Foster, 2011).
Spatial and relational comparisons. A second, more powerful idea is that the visual system compares surfaces with one another rather than measuring each in isolation. Land and McCann's retinex theory (from retina + cortex) proposed that the visual system computes, separately within each waveband, the ratios of light reflected from neighboring regions across the scene; because the illuminant scales all surfaces together, these spatial ratios are largely invariant to the light and track relative reflectance (Land & McCann, 1971). There is a precise physical reason this works: the spatial ratios of cone excitations between pairs of surfaces stay very nearly invariant when the illuminant changes, across large collections of natural surfaces and daylights — so a visual system that reads those ratios gets stability almost for free (Foster & Nascimento, 1994). Relational comparison — color computed across space, not point by point — remains central to how the field thinks about constancy.
Cues to the illuminant. Real scenes are full of hints about the light, and the visual system appears to use them: the color of specular highlights (which carry the illuminant's color directly), the brightest or most diffusely reflective surface taken as a reference for "white," mutual reflections between surfaces, and the average color of the whole image. Each cue is a partial solution, and constancy improves as more of them are present (Foster, 2011; Witzel & Gegenfurtner, 2018).
How good are these mechanisms? A landmark experiment tested whether constancy could be explained by the three simplest adaptation rules — adaptation to the local surround, to the spatial mean of the image, or to the most intense region. By varying the scene, Kraft and Brainard titrated constancy across a wide range and showed that none of the three simple rules could account for the results: real constancy under near-natural viewing draws on more than any single low-level adaptation mechanism (Kraft & Brainard, 1999).
Computational and Bayesian models. Formal theories make the inference explicit. Maloney and Wandell showed that if natural surface reflectances and illuminants can each be approximated by a small number of basis functions — a "linear model" with about three parameters — then a visual system can, in principle, recover surface reflectance from cone signals despite unknown illumination (Maloney & Wandell, 1986). Brainard and Freeman recast the whole problem in the language of Bayesian decision theory: combine prior probability distributions over likely illuminants and surfaces with the incoming photoreceptor data, then choose the most probable interpretation. They argued that the right estimator for perception is not the single most likely answer but one that finds the most probable approximately correct answer, and showed this reproduces aspects of human performance (Brainard & Freeman, 1997). On this view, Helmholtz's "unconscious inference" is literally a computation of posterior probability.
Simultaneous Color Contrast: The Flip Side of Constancy
If the visual system judges color by comparing each surface with its neighbors, then changing a surface's surroundings should change how it looks even when the surface itself is unchanged. It does — an effect called simultaneous color contrast (chromatic induction): a neutral gray patch looks faintly greenish on a red background and faintly reddish on a green one, because the surround induces its opponent color into the patch.
Contrast and constancy are two faces of the same strategy. By coding color relative to the surround, the visual system gains stability against changes in the overall light (constancy) at the cost of some accuracy about any single patch in isolation (contrast). The canonical neural substrate for these spatial comparisons is the double-opponent cell, which is excited by one color in its center and the opposite color in its surround (Shapley & Hawken, 2011).
The Neural Basis
Where in the brain does color become constant? Recording from single cells in the monkey visual cortex, Zeki drew a crucial distinction between two kinds of color-related neuron. Wavelength-selective cells, common in the primary visual cortex, respond to the wavelength composition of the light in their receptive field — they track the physical light, not the perceived color. But in the V4 complex he found cells that respond to a surface's natural color — its perceived color when viewed as part of a full scene — largely independent of the exact wavelengths reaching the eye, while ignoring its "void" color seen in isolation (Zeki, 1983). This made area V4 a landmark in the cortical construction of color.
The modern picture is more distributed and less tidy. Double-opponent cells in the primary visual cortex (V1) and V2 — cells with spatially opposed, chromatically opponent center and surround — provide exactly the spatial color comparisons that relational constancy requires, and the role of V1 in color is now seen as much larger than the older "color center in V4" story implied (Shapley & Hawken, 2011). Rather than a single locus, color constancy appears to be computed in stages along the ventral stream, with the exact division of labor still an open question (Witzel & Gegenfurtner, 2018).
When Constancy Breaks Down
Color constancy is partial and defeasible, and its failures are as instructive as its successes.
"The dress." In 2015 a single photograph of a dress went viral because people sharply disagreed about its colors: some saw white and gold, others blue and black. Tested formally in over 1,400 people, the split was real and stable, the largest documented individual difference in color perception — and, tellingly, it mapped onto age and sex (Lafer-Sousa, Hermann, & Conway, 2015). The leading explanation is a color-constancy account: the image is ambiguous about its illumination, and observers who implicitly assume the dress is lit by cool bluish daylight discount blue and see white/gold, while those who assume warm indoor light discount orange and see blue/black (Brainard & Hurlbert, 2015). The dress is striking precisely because it exposes the normally invisible inference at the heart of constancy: when the scene fails to pin down the illuminant, different people's priors fill the gap differently, and the same pixels become different colors. Exactly why observers differ is still debated.
Monochromatic light. Some illuminants defeat constancy by physics alone. Under a low-pressure sodium street lamp, which emits essentially a single wavelength, there is almost no spectral information for the visual system to exploit — surfaces differ only in how much of that one wavelength they reflect — and color vision collapses into shades of yellow-gray. No amount of clever inference can recover what the light never carried (Smithson, 2005).
Metamer mismatching. A subtler failure happens under broadband but spiky lights such as fluorescent and LED lamps. Two surfaces that reflect different spectra can match under one light yet visibly differ under another — illuminant-induced metamer mismatching. This is why a shirt and trousers that look like a perfect match in a store can clash in daylight; the frequency and size of such mismatches have been measured directly across natural scenes (Foster et al., 2006).
Memory and Knowledge
Constancy is not purely a matter of light and surfaces; what you know about an object can tint what you see. In a now-classic demonstration, observers adjusted images of familiar fruit until they looked gray. They overshot in a revealing way: to make a banana appear neutral gray, people had to make it slightly blue — the opponent of yellow — because their knowledge that bananas are yellow was adding yellow to the percept. At the point where the banana was physically achromatic, it still looked faintly yellow (Hansen, Olkkonen, Walter, & Gegenfurtner, 2006). Follow-up work showed that such memory color effects persist across changes in simulated illumination, consistent with object knowledge acting as one more cue that helps stabilize color (Olkkonen, Hansen, & Gegenfurtner, 2008). These effects place a genuinely cognitive component alongside the sensory and computational ones (Smithson, 2005).
A Worked Example: A White Mug, Morning and Night
Picture a plain white mug on your desk. In the morning, sunlight through the window is bluish, so the light the mug reflects into your eye is, physically, bluish-white. At night under a warm lamp, the very same mug reflects orange-white light. The raw cone signals in the two cases are different colors — yet you see one white mug all day.
Here is what the layered machinery is doing. Adaptation rescales your cone gains to the prevailing light, partly cancelling the overall color cast (Kries, 1905/1970). Relational comparison helps too: the mug is the brightest, most neutral thing on a desk of colored objects, and its cone-excitation ratios to those objects barely change when the light does (Foster & Nascimento, 1994). The highlight glinting off the mug's glaze carries the lamp's color directly, giving the visual system a read on the illuminant (Witzel & Gegenfurtner, 2018). Your visual system combines these into an estimate of the light and discounts it (Brainard & Freeman, 1997). And because you know the mug is white, memory color nudges the percept the rest of the way (Hansen et al., 2006). What feels like "just seeing a white mug" is in fact a continuous inference about what the light is doing.
Color Constancy Compared With Related Ideas
Color constancy is one member of a family of stabilizing achievements and is easily confused with the lower-level processes that contribute to it.
| Concept | What it refers to | Relation to color constancy |
|---|---|---|
| Color constancy | Stable perceived surface color across changes in illumination | The phenomenon itself |
| Chromatic adaptation | Receptoral rescaling of cone gains to the prevailing light | A low-level mechanism that contributes to constancy, but not the whole of it |
| Simultaneous color contrast | A surface's apparent color shifts with its surround | The relational coding that underlies constancy, seen from the other side |
| Lightness constancy | Stable perceived surface lightness (achromatic) across illumination | The achromatic analogue; the same inverse problem for intensity |
| Perceptual constancy | Stable perceived size, shape, lightness, and color of objects | The general family color constancy belongs to |
| Color appearance | How a color actually looks under given viewing conditions | What constancy stabilizes; modeled by color-appearance models |
Contemporary Research
Modern work pushes in several directions. The dominant modeling framework is the equivalent-illuminant approach, which fits human color matches by assuming the visual system behaves as if it estimates a single effective illuminant and discounts it, then asks how scene information sets that estimate (Brainard & Maloney, 2011; Brainard & Freeman, 1997). A second strand is natural-scene statistics: large hyperspectral datasets of measured surfaces and daylights make it possible to ask what assumptions are actually warranted by the world, and to test whether invariant image features can support constancy without explicitly estimating the light (Foster, 2011; Foster et al., 2006). A third is machine color constancy, where the same problem — recover surface color, or "white balance" an image, under unknown illumination — drives computer-vision systems, with the linear-model and Bayesian formulations from vision science as the conceptual backbone (Maloney & Wandell, 1986). Color categories add a further twist: how constancy interacts with the named colors people use is an active question (Witzel & Gegenfurtner, 2018).
Criticisms and Open Questions
The field's hardest question may be how much "color constancy," classically defined, really exists. Foster argued that measured constancy for the abstract goal of recovering surface reflectance is often only moderate, and that what is robust is relational color constancy — the ability to tell that two surfaces differ, or that a change was due to the light rather than the material — which the visual system performs very well (Foster, 2011; Foster & Nascimento, 1994). This reframes constancy less as perfect reflectance recovery and more as a suite of related, partial competencies. Three open issues stand out: the degree of constancy is highly sensitive to the scene and the experimental task, so single numbers mislead (Kraft & Brainard, 1999); individual differences can be large, as the dress made vivid, and we do not fully understand where people's differing illuminant priors come from (Lafer-Sousa et al., 2015); and the relative weight of low-level sensory versus high-level cognitive contributions remains contested (Smithson, 2005; Hansen et al., 2006).
Key Researchers
Hermann von Helmholtz (1821–1894) — Founder of the inferential view of color; argued that the visual system discounts the illuminant through unconscious inference to recover stable surface color (Helmholtz, 1867/1924).
Johannes von Kries (1853–1928) — Formulated the coefficient law of chromatic adaptation — independent gain control in the three cone channels — still the textbook first approximation to color constancy (Kries, 1905/1970).
Edwin H. Land (1909–1991) — With John J. McCann, formulated retinex theory, computing color from spatial reflectance ratios across the scene, and demonstrated it with Mondrian displays (Land & McCann, 1971).
Semir Zeki — University College London; Emeritus Professor of Neuroaesthetics; identified cells in cortical area V4 that track a surface's stable "natural" color rather than the wavelengths reaching the eye, a neural correlate of constancy (Zeki, 1983). Faculty
David H. Brainard — University of Pennsylvania; RRL Professor of Psychology; developed Bayesian and equivalent-illuminant models of color constancy and rigorous psychophysical tests of its mechanisms (Brainard & Freeman, 1997; Brainard & Maloney, 2011). Faculty
David H. Foster — University of Manchester; provided a comprehensive synthesis of the measurement and theory of color constancy and established the cone-excitation-ratio basis of relational color constancy (Foster, 2011; Foster & Nascimento, 1994). Faculty
Key Terms
| Term | Meaning |
|---|---|
| Color constancy | The tendency for perceived surface color to remain stable despite changes in the illumination. |
| Surface reflectance | The fraction of light a surface reflects at each wavelength — a stable property of the object. |
| Illuminant | The light falling on a scene, described by its spectral power distribution. |
| Color signal | The light reaching the eye from a surface — the product of illuminant and reflectance. |
| Inverse problem | Recovering surface reflectance from the color signal, when the illuminant is unknown; mathematically underdetermined. |
| Discounting the illuminant | Estimating the light and removing its contribution to recover the surface's own color. |
| Constancy index | A 0-to-1 measure of how completely perception tracks surface reflectance (1) versus the raw light (0). |
| Chromatic adaptation | Receptoral rescaling of cone responses to the prevailing light (the von Kries coefficient law). |
| Retinex theory | Land and McCann's account in which color is computed from spatial reflectance ratios across the scene. |
| Cone-excitation ratio | The ratio of a cone class's response to one surface versus another; nearly invariant under illuminant changes. |
| Relational color constancy | The robust ability to judge that surfaces differ, or that a change was due to the light, rather than recover absolute reflectance. |
| Simultaneous color contrast | The shift in a surface's apparent color caused by the color of its surround (chromatic induction). |
| Double-opponent cell | A cortical neuron with chromatically opponent, spatially opposed center and surround — a substrate for spatial color comparison. |
| Metamerism / metamer mismatch | When two surfaces match under one illuminant but differ under another. |
| Memory color | The influence of stored knowledge of an object's typical color (e.g., bananas are yellow) on its perceived color. |
| Equivalent illuminant | A single effective illuminant the visual system behaves as if it estimates and discounts. |
Frequently Asked Questions
What is color constancy in simple terms?
It is the reason objects keep looking the same color under very different lights — a white shirt looks white in bluish daylight and under a warm lamp, even though the light it reflects is physically bluish in one case and orange in the other. The visual system estimates the illumination and discounts it to recover the object's own color (Foster, 2011).
Why is color constancy considered a hard problem?
Because the light reaching the eye is the product of the illumination and the surface reflectance, and the eye cannot measure them separately. Many light-and-surface combinations produce the same signal, so the problem is mathematically underdetermined and can be solved only with added assumptions (Smithson, 2005).
How does the brain achieve color constancy?
Through layered mechanisms: chromatic adaptation rescales the cone responses to the prevailing light; the visual system compares reflectance ratios across surfaces (retinex); it reads cues to the illuminant such as highlights and a white reference; and it applies prior knowledge, including memory color. These combine into an estimate of the light, which is discounted (Land & McCann, 1971; Kraft & Brainard, 1999).
Is color constancy perfect?
No. It is usually partial, and how complete it is depends on the scene; in experiments the degree of constancy can be moved from about 10% to over 80% by changing how much information the scene provides (Kraft & Brainard, 1999; Foster, 2011).
Is camera white balance the same as color constancy?
They solve the same problem. A camera's white balance estimates the color of the light and corrects the image so that whites look white — an engineered version of what the visual system does automatically. Auto white balance can fail in the same situations that fool human constancy, such as a scene lit by a single strong color (Maloney & Wandell, 1986; Brainard & Maloney, 2011).
What does "the dress" tell us about color constancy?
The dress photograph is ambiguous about its illumination, so observers who implicitly assume bluish daylight discount blue and see white/gold, while those who assume warm light discount orange and see blue/black. It is a vivid case of color constancy working differently across people (Lafer-Sousa, Hermann, & Conway, 2015; Brainard & Hurlbert, 2015).
Where in the brain does color constancy happen?
Wavelength-selective cells in early visual cortex track the physical light, but cells in cortical area V4 respond to a surface's stable perceived ("natural") color; double-opponent cells in V1 and V2 supply the spatial color comparisons constancy relies on. The computation is distributed rather than confined to one area (Zeki, 1983; Shapley & Hawken, 2011).
Does what I know about an object change its color?
Yes, to a measurable degree. Because people know bananas are yellow, a physically gray banana looks faintly yellow, and observers must make it slightly blue to see it as neutral — a "memory color" effect that holds up across changes in lighting (Hansen, Olkkonen, Walter, & Gegenfurtner, 2006; Olkkonen, Hansen, & Gegenfurtner, 2008).
References
| 1 | Brainard, D. H., & Freeman, W. T. (1997). Bayesian color constancy. Journal of the Optical Society of America A, 14(7), 1393–1411. https://doi.org/10.1364/JOSAA.14.001393 |
| 2 | Brainard, D. H., & Hurlbert, A. C. (2015). Colour vision: Understanding #TheDress. Current Biology, 25(13), R551–R554. https://doi.org/10.1016/j.cub.2015.05.020 |
| 3 | Brainard, D. H., & Maloney, L. T. (2011). Surface color perception and equivalent illumination models. Journal of Vision, 11(5):1, 1–18. https://doi.org/10.1167/11.5.1 |
| 4 | Foster, D. H. (2011). Color constancy. Vision Research, 51(7), 674–700. https://doi.org/10.1016/j.visres.2010.09.006 |
| 5 | Foster, D. H., Amano, K., Nascimento, S. M. C., & Foster, M. J. (2006). Frequency of metamerism in natural scenes. Journal of the Optical Society of America A, 23(10), 2359–2372. https://doi.org/10.1364/JOSAA.23.002359 |
| 6 | Foster, D. H., & Nascimento, S. M. C. (1994). Relational colour constancy from invariant cone-excitation ratios. Proceedings of the Royal Society B: Biological Sciences, 257(1349), 115–121. https://doi.org/10.1098/rspb.1994.0103 |
| 7 | Hansen, T., Olkkonen, M., Walter, S., & Gegenfurtner, K. R. (2006). Memory modulates color appearance. Nature Neuroscience, 9(11), 1367–1368. https://doi.org/10.1038/nn1794 |
| 8 | Helmholtz, H. von. (1924). Helmholtz's treatise on physiological optics (Vol. 2; J. P. C. Southall, Ed. & Trans.). Optical Society of America. (Original work published 1867) |
| 9 | Kraft, J. M., & Brainard, D. H. (1999). Mechanisms of color constancy under nearly natural viewing. Proceedings of the National Academy of Sciences, 96(1), 307–312. https://doi.org/10.1073/pnas.96.1.307 |
| 10 | Kries, J. von. (1970). Influence of adaptation on the effects produced by luminous stimuli. In D. L. MacAdam (Ed.), Sources of color science (pp. 109–119). MIT Press. (Original work published 1905) |
| 11 | Lafer-Sousa, R., Hermann, K. L., & Conway, B. R. (2015). Striking individual differences in color perception uncovered by 'the dress' photograph. Current Biology, 25(13), R545–R546. https://doi.org/10.1016/j.cub.2015.04.053 |
| 12 | Land, E. H., & McCann, J. J. (1971). Lightness and retinex theory. Journal of the Optical Society of America, 61(1), 1–11. https://doi.org/10.1364/JOSA.61.000001 |
| 13 | Maloney, L. T., & Wandell, B. A. (1986). Color constancy: A method for recovering surface spectral reflectance. Journal of the Optical Society of America A, 3(1), 29–33. https://doi.org/10.1364/JOSAA.3.000029 |
| 14 | Olkkonen, M., Hansen, T., & Gegenfurtner, K. R. (2008). Color appearance of familiar objects: Effects of object shape, texture, and illumination changes. Journal of Vision, 8(5):13, 1–16. https://doi.org/10.1167/8.5.13 |
| 15 | Shapley, R., & Hawken, M. J. (2011). Color in the cortex: Single- and double-opponent cells. Vision Research, 51(7), 701–717. https://doi.org/10.1016/j.visres.2011.02.012 |
| 16 | Smithson, H. E. (2005). Sensory, computational and cognitive components of human colour constancy. Philosophical Transactions of the Royal Society B: Biological Sciences, 360(1458), 1329–1346. https://doi.org/10.1098/rstb.2005.1633 |
| 17 | Witzel, C., & Gegenfurtner, K. R. (2018). Color perception: Objects, constancy, and categories. Annual Review of Vision Science, 4, 475–499. https://doi.org/10.1146/annurev-vision-091517-034231 |
| 18 | Zeki, S. (1983). Colour coding in the cerebral cortex: The reaction of cells in monkey visual cortex to wavelengths and colours. Neuroscience, 9(4), 741–765. https://doi.org/10.1016/0306-4522(83)90265-8 |