What is so pleasant about the cerulean of a clear summer sky? Or the verdure of a freshly trimmed lawn? The diversity of words for color testifies to our repeated attempts to describe them. Sometimes it feels like even superlatives – the “bluest” sky or the “greenest” grass – are not enough to represent the sensation of color as a mere word. Despite the many words describing colour, colour itself is not so easy to define. What exactly is colour?

Colour is not a property of the environment. It is not, for instance, defined by the wavelength of the light that an object reflects or emits onto the retina. Consider, for example, a yellow shirt that appears yellow because it reflects light near 590 nm in its wavelength. Now consider an RGB display showing a picture of the same shirt. What colour is the shirt on the display? Any reasonable observer would agree this shirt is also yellow, even though the pixels are only emitting the wavelengths around 500 nm and 700 nm that we traditionally call “red” and “green.” So colour cannot be synonymous with wavelength.

Colour is also not an objective mapping from observed spectra to perception. For example, we agree that some color vision deficient people cannot see the bright reds of a sunset – particularly protanopes, who lack the photoreceptors corresponding to that range of wavelengths. If the observer does not see a red sunset, is the sunset still red? Beyond protanopes, there exists a variety of colour vision-related mutations and learned associations that all change people’s experience of colour, making an objective mapping from spectrum to colour unrealistic.

Instead, colour refers to a subjective experience, in the same way consciousness and subjective time are subjective experiences. Consider the question: how do I know if anyone else sees the same red that I do? If we recognize colour as a subjective experience, answering this question is as hard as answering whether someone else is conscious. Even knowing that someone’s neural responses are the same would not imply that they have the same subjective experience of that colour.

Colour, then, is not an external property but a construction. What is the shape of this construction, and what is its purpose? Knowing the answers to these questions could explain much about how we see colour.

For example, why does a combination of red and green light appear yellow, when neither contains a yellow wavelength? The answer lies in our form of colour vision, which is called trichromacy. In trichromacy, colour vision involves three kinds of cone cells in the retina, which each contain photoreceptors sensitive to a different range of wavelengths. The three types of cones are S, M, and L cones, which correspond roughly to blue, green, and red wavelengths respectively. These cone cells fire with a certain probability in response to each wavelength of light, with the highest probability of firing at the target wavelength and a lower probability of firing at the surrounding wavelengths.

The wavelength ranges to which cones are sensitive overlap, and their overlapping ranges are key to the accurate perception of colour. Even though a cone is expected to detect both the brightness and wavelength of an incoming stimulus, it is univariate – since it’s firing rate only represents values along one dimension – which means that it cannot disambiguate the dimensions of wavelength and brightness. That’s why the cones have overlapping ranges of sensitivity – the original wavelengths of incoming light can be deduced based on the ratio of the outputs from the three different kinds of cone cells.

Trichromacy explains quite well why a screen with red and green pixels looks yellow. Given that the experience of yellow light corresponds to the firing of (mostly) M-cones and L-cones, a screen can simulate yellow light by directly adjusting the intensities of red and green – without needing to produce any light of a “yellow” wavelength. The output hue would be completely indistinguishable from that of a true yellow wavelength. Indeed, colour displays are a primitive but powerful example of virtual reality based on hijacking the brain’s encoding mechanisms – using reverse engineered principles of trichromatic vision to simulate colours that aren’t there.

What trichromacy doesn’t quite explain is how certain pairs of colours preclude each other from being seen. Why, for example, do blue and yellow light together look like white? As a related question, why do you see purple afterimages after looking at a bright sunny scene? These effects are grouped by vision scientists under the label of “color opponency.”

An early theory explaining opponency was Hering’s Opponent Process Theory, which states that colors of “unique hues” (red, green, yellow, blue) inhibit the observation of their opposite hues. Some early studies by Russell de Valois, David Hubel, and Torsten Wiesel, seemed to identify physiological basis for colour opponency. They proposed opponency working in two stages, through (retinal) single opponent and (cortical) double opponent cells. Single opponent cells are ganglion cells in the retina, which either select for red points surrounded by green backgrounds, green points surrounded by red backgrounds, yellow surrounded by blue, or blue surrounded by yellow. Because these ganglion cells (and their corresponding lateral geniculate nucleus cells) do not include the “opponent” cone type inputs in their summations, they are just as sensitive to a white light shined in the center as they are sensitive to their original colour. Meanwhile, double opponent cells are higher-level visual cortex cells that do differentiate between these two cases, by treating red inputs as excitatory and green inputs as inhibitory in the center of their receptive field (and so on for the three other cases).

More recently, this physiological evidence has been called into question. As Solomon and Lennie point out in their 2007 review, there is little evidence of the clean-cut distinction between different kinds of opponent cells in the visual cortex. Although Hering’s Opponent Process theory sounds like a convincing explanation for colour opponency, it doesn’t quite align with physiological reality. In general, colour opponency is still poorly understood, with no accepted theory accounting for this fundamental fact about colour appearance.

Another confounding factor in the construction of colour is that illumination cannot be disentangled from the reflected light. For example, a red tomato might look green in a green light, because there is no red light available to be reflected. Despite the variety of natural illumination sources, however, we generally perceive a remarkable consistency in the colour of objects. Small changes in illumination, such as a switch from sunlight to skylight, do not affect our perception of a tomato being red. One way this might happen – as proposed by Polaroid founder Edwin Land in his Retinex theory – is that the outputs of opponent processes are normalized according to the average value in the image – this should cancel out the effects of illumination as long as the illuminant contains relatively similar amounts of each opposing component.

While Land’s theory was a good approximation for human colour constancy, further research has shown that real-life colour constancy is more complicated, involving not just hues but also cues like shadow, brightness, and even texture – for example, wet rock looks different from dry rock. How do we differentiate wet rock from dark rock and infer the colour of the concrete below? So, it seems that a satisfactory theory for colour constancy would also have to explain the broader phenomenon of perceptual constancy, because colour is often inseparable from features like texture and lightness in the real world.

Beyond its biological basis, however, it’s useful to consider why colour vision might matter to humans. Perhaps the most important and widely accepted explanation is that colour vision plays an important role in differentiating adjacent objects (e.g. telling a mango apart from a mango tree leaf), which might be hard to differentiate in the absence of colour cues. In support of this theory, evolutionarily relevant colours like the redness of a fruit (corresponding to ripeness) and the blue of a clear lake appear especially salient to human observers. Indeed, in some cases – as with red fruit – it’s likely that the colour of the fruit evolved specifically to allow easier detection by its consumers, which were presumably animals with colour vision.

This idea – that our perception of colour can shape the development of the outside world – complicates our earlier definition of colour. Though colour is supposedly a subjective internal experience, it has also irrevocably changed our external world. Walk into any supermarket, and you are confronted with a barrage of artificial colours adorning advertisements, packaging, and even the clothing of other shoppers. In pursuit of these colours, people have done everything from plucking small insects from cacti for carmine red to squeezing the mucus out of thousands of sea snails for Tyrian purple.

In other words, even as our perception of colour is a construct that helps us survive in the world, the world itself is again modified by our existence as beings that perceive colour. In a sense, then, colour really is an observable property of the environment, but only if we shift to an understanding of the environment as a consequence of human behavior.

References

Hurlbert A (1997) “Colour Vision (Primer).” Current Biology. 7(7): R400-402. https://www.cell.com/current-biology/fulltext/S0960-9822(06)00199-0

Gouras, P. Colour vision. https://webvision.med.utah.edu/book/part-vii-color-vision/color-vision/

Bowmaker, J. K. (1983). Trichromatic colour vision: why only three receptor channels?. Trends in Neurosciences, 6, 41-43.

Mollon, J. D. (1989). “Tho’she kneel’d in that place where they grew…” The uses and origins of primate colour vision. Journal of Experimental Biology, 146(1), 21-38.

Smithson HE (2005) “Sensory, computational and cognitive components of human colour constancy.” Philosophical Transactions of the Royal Society: Biological Sciences. 360 (1458): 1329-1346.

Solomon SG & Lennie P (2007) “The machinery of colour vision”. Nature Reviews Neuroscience, 8: 276-286.