Posted 5 March 2018. Last updated 25 Feb 2023.
A letter carrying the gist of this essay was published in New Scientist, 3150, 4 November 2017. Online subscribers can find it here.
When I first came across fictional descriptions of colours which do not exist in normal life (possibly by H P Lovecraft?), I wondered what the author was getting at. Later I discovered that people who experience visual hallucinations, typically after taking a psychedelic drug, sometimes report such colours. A friend of mine was once, for other reasons, prescribed a drug whose side effects turned out to include hallucinations. While watching a painting of a waterfall cascade down all over the sofa and floor, he also told me that he could see colours which he had never seen before. He was not colour blind, these were genuinely new colours. Sadly, I was refused a pill to try it for myself, but I wondered where on earth such colours could come from. Now, I think may know.
It is well known that our eyes see in red, green and blue, and these RGB colour signals are then combined in the brain to make every possible visible shade. TV and phone screens typically work on the same principle. The basic spectrum of colours is often represented on a colour wheel, while the full array of shades – the "colour space" – which a screen creates is represented in a colour cube. Each dimension represents one of the primary colours of vision. One long diagonal axis of the cube is occupied by the core black-to-white (K to W) monochrome shades of grey. Moving outwards away from this core brings ever more intense colours. Above black, the nearest three corners are the red, green and blue primary colours seen by the eye. Above these lie the cyan, magenta and yellow of the colour printer, with each of these being just two of the primaries added together. For example yellow is just red added to green.
The human eye is not a perfect receiver; among other things the three "cone" types of colour sensors are pretty rough-and-ready and some extra pre-processing has to be done to draw out the full RGB palette. As a result the colours which we actually perceive form a colour space more like a lopsided double-cone. But it includes much the same colours and the same principle apply, so the cube representation is good enough for the present purpose – and a lot simpler for me to draw.
But this simple colour-space model does not explain some of the more subtle effects of vision, such as why the colours in after-images appear as they do – or how nonexistent colours might appear. The way the brain processes colour is indeed a bit more subtle.
According to the standard opponent-process theory of colour vision, one of the first things that the brain does to colour signals from the eye is to convert the three red, green and blue signals down to just two; a red-green difference signal and a blue-yellow difference. Among other things this model is said to explain the colours of after-images, a distinct improvement on the RGB model.
The exact conversion scheme used is open to debate. Pridmore has recently argued that the scheme traditionally assumed is flawed, and that the neurology is better explained by a modified scheme. In particular, the conventionally described "red" cone cells in the eye are actually producing the yellow signal; the red component is created by adding the yellow and blue cone signals together. In my graphics I depict the standard scheme, but either way it makes little difference to the thrust of my argument.
I suggest here that the opponent-process model can also account for seeing non-existent colours. (if this has been proposed before, I have never come across it. Please tell me if it has, I'd love to read about it!)
Just as the colour cube describes the colour space of RGB vision, so too we can define an opponent colour space. It is again three-dimensional, with one dimension for the red-vs-green signal and another for the blue-vs-yellow mixdown. The third dimension is provided by the monochrome black-to-white brightness level or greyscale, which is produced by a different kind of "rod" light sensor in the eye. I have left it out until now for simplicity's sake.
The RGB colour space must fit inside this opponent colour space, or we would not be able to perceive all the colours seen by the eye. Crucially, the fit is not exact and the opponent space is actually larger than the RGB space. The diagram shows how the RGB cube fits inside. If we leave the cube undistorted (it should really be squished into that lopsided double-cone), then the opponent space is itself not quite a cube. The RGB space touches it at its own black, white, yellow and blue corners (shown by black dots). It also touches along its green-cyan and red-magenta edges (shown by a thick line), for example the RGB red-magenta edge is also an opponent red-but-no-green edge. The regions outside the RGB space can, by definition, never be stimulated by the eye, so most of us live out our lives blissfully unaware of them.
But what happens if you over-excite this area of the brain, perhaps through some genetic or traumatic upset, or the administration of a hallucinogenic drug? In this dysfunctional state, perhaps it could be stimulated to produce signals in this larger colour space outside of normal vision, and to pass them on to the conscious areas of the brain. These novel visual signals could (and, I suggest, most likely would) be perceived as colours never normally seen, exactly what my friend reported.
So there you have it, my take on the way in which people can sometimes see colours which do not, and never can, exist in the real world.
Intriguingly, the old UK analogue TV broadcasts used a similar kind of conversion down to two colour channels, in order to reduce the bandwidth for transmission. The human brain may have evolved the same basic technique for much the same reasons.
The PAL system used something called Y'UV colour encoding to minimise its demands on bandwidth. Y' is the luminance corresponding to a black-and-white transmission (fairly close to, but not quite the same as, a pure yellow signal). Although the Y' signal is created directly by black-and-white cameras, for colour transmission it is a complex blend of the R, G and B signals produced by the camera. U and V are different blends of the primary colour channels, designed to minimise data transmission bandwidth by matching human sensitivities (U is blue-monochrome difference, V is red-monochrome difference, where the monochrome signal is the already-complex Y').
The human eye is a little different. Besides the three colour channels from the cone cells, it also produces a direct monochrome signal from its rod cells like a black-and-white camera. The eye thus creates four signals, in effect R, G, B and Y'.
Despite their subtly different inputs (Y'UV requires only RGB) both Y'UV and the opponent-process system reduce the colour palette down to a monochrome signal plus two difference mixes, in such a way that it reduces bandwidth yet can be unscrambled at the other end. It seems to me that if the opponent model is correct then this could hardly be a coincidence: as happens so often in technology, once nature had evolved an elegant and effective way of highlighting the information that mattered most, TV engineers found themselves with the same goal, to develop a practical image encoding for minimal effort. It is remarkable how close their solution is to Nature's, despite not yet fully understanding how she had done it.
There is one other thing in all this which still puzzles me. A colour wheel of the opponent-process model can be made, analogous to the RGB one. In it, you can see that red and yellow are separated apart, making room for orange, and lime is similarly emphasised between yellow and green. Everything else gets a bit squashed up.
The appearance of orange is closer to our conscious perception and results in a more even distribution of the colours we see around the wheel. But on the other hand, lime green also appears and that is not a colour we perceive as primary – any more than cyan is.
A typical colour wheel reflecting the way we consciously perceive colour is based on red, yellow and blue. This intuitive RYB breakdown was in fact the first one proposed. The prominence of orange is notable, while cyan and lime, on either side of green, are no longer emphasised. Is there perhaps a further level of hard-wired processing in which the colours are mixed around once more, to produce this particular distribution of perceived colours? (I once read somewhere that there was evidence for this, but the source gave no reference. Please contact me if you can help.)
One way and another the brain has a fair problem to deal with. In the physical colour spectrum, the colours that we have found most useful (and we have therefore evolved to be most sensitive to) are mostly grouped at one end of the visible spectrum, around the yellow which is the Sun's strongest light. Unravelling that spectrum to produce the perception of evenly-spaced red, orange, yellow, green and blue, and even overlapping the red and blue to make the violet which completes the circle, is no mean challenge. Violet itself only appears as a kind of double-resonance of red, where the most energetic blue light approaches exactly twice the energy of red light, allowing a small number of "double-hits" on a red cone to mix with the blue signal as our vision tails off into the ultra-violet. But that does not reach round to the purple-reds between pure red and violet, these shades must always be made up of impure light containing multiple colours and our brains then invent the apparent purity of the colours which close the circle.
Moreover, the RGB spectrum is unlikely to be the last word in our evolutionary development. Perhaps at one time or another, our ancestors could see further into the infrared and/or ultraviolet, as many other creatures can today. Even though those receptors would have disappeared from our eyes, perhaps the neural infrastructure is still in place, to be stimulated by drugs or trauma and create false images in colours we have long forgotten? This could offer an alternative route to psychedelia.
The many languages we use around the world have a remarkable variety of ways to express colours, often using words to group shades quite differently from the way the major European languages do. Even in the West, with identical eye pigments, a colour firmly described by one person as "blue" can be equally firmly described by another as "green". Might the perceptual level of processing be to some extent cultural, learned from those around us during our early childhood, as we acquire language and associate specific words with specific colour ranges? Might it then become instinctive, like riding a bicycle? Perhaps colour processing is not hard-wired by evolution after all. Or, might both extra layers of processing be present, might language and culture represent yet another layer of "soft" perceptual processing on top of the hard-wired layers? I would love to know.
Pridmore, R.W. "A new transformation of cone responses to opponent color responses". Attention, Perception and Psychophysics 83, 1797–1803 (2021). DOI 10.3758/s13414-020-02216-7