Colour Vision: How the Eye Works, and Colour Blindness

Colour is a sensation, not a property of things

It is tempting to say that a ripe tomato is red, in the way that it is round and weighs ninety grams. It is not. The tomato reflects a particular mixture of wavelengths, and a visual system built from three kinds of light-sensitive cell, wired into a brain that has strong opinions about context, produces the experience we call red. Change the observer and the experience changes. Roughly eight per cent of men of northern European descent would describe that tomato quite differently against its leaves, and a mantis shrimp would presumably find our entire account of the matter charmingly impoverished.

For anyone working with colour professionally, this is not philosophy. It is the reason a colour that measures identically on two samples can look different to two people, and the reason a chart that reads beautifully to a designer can be unreadable to a tenth of the audience. Understanding the biology of colour vision is the foundation for everything else — measurement standards, screen calibration, palette design, accessibility. This article covers how the eye actually turns light into colour, where the system is fragile, and what colour vision deficiency really involves.

The retina: two systems doing different jobs

Light entering the eye is focused onto the retina, a thin sheet of neural tissue lining the back of the eyeball. Two families of photoreceptor cells sit there, and they do not do the same work.

Rods number around 120 million. They are extraordinarily sensitive — a well-adapted rod can respond to a single photon — but they come in only one flavour. A system with one receptor type cannot distinguish colour, because it cannot tell a dim short-wavelength light from a brighter long-wavelength one. This is why the world goes grey at night. The phenomenon has a name: scotopic vision, the low-light regime in which rods dominate and colour discrimination effectively vanishes.

Cones number around six million and need far more light to respond. They come in three types, and it is the comparison between their responses that creates colour. In bright conditions — photopic vision — cones do essentially all the work. In the intermediate mesopic range, both systems contribute, which is one reason colour judgements made in dim showrooms or at dusk are unreliable.

Why the cones are not “red, green and blue”

The three cone types are properly called L, M and S, for long, medium and short wavelength sensitivity. Their peak sensitivities sit at roughly 560, 530 and 420 nanometres. Note how close L and M are — only about thirty nanometres apart, both peaking in what we would call the yellow-green region. They are not tuned to red and green at all; they overlap heavily, and the visual system extracts colour from the small difference between two broadly similar signals.

This has a practical consequence that surprises people. Our finest colour discrimination is not in the reds or blues but in the yellow-green band where L and M responses diverge most sharply per unit of wavelength change. In the deep blues and deep reds, discrimination is comparatively coarse. It also explains why the S cones, which are sparse, absent from the very centre of the fovea, and slow, contribute little to fine detail. Blue text on a dark ground looks soft and unfocused partly for this reason.

From three signals to opponent channels

The retina does not simply forward three cone signals to the brain. Before the information leaves the eye, it is recoded into three opponent channels: a light-dark (achromatic) channel, a red-green channel built mainly from L minus M, and a blue-yellow channel built from S minus the sum of L and M.

Opponent coding explains several everyday observations that a simple three-channel model cannot.

  • Impossible colours. You can imagine a reddish yellow (orange) or a bluish green (teal), but not a reddish green or a yellowish blue. Those pairs sit at opposite ends of the same channel, so a single channel cannot signal both at once.
  • Afterimages. Stare at a saturated red patch for thirty seconds and look at white paper: a green ghost appears. The red-green channel has adapted, and its return to neutral is read as a signal in the opposite direction.
  • Unique hues. Most observers can point to a green that is neither yellowish nor bluish, and a blue that is neither greenish nor reddish. These are the channel null points, and they are remarkably consistent across cultures.
  • Efficient transmission. The achromatic channel carries most of the spatial detail; the chromatic channels carry far less. Image and video compression exploits this directly through chroma subsampling, throwing away colour resolution that the eye was never going to use.

The brain edits aggressively

Between the retina and conscious perception, the signal is heavily processed, and the processing is not neutral. Two mechanisms matter most for colour work.

Chromatic adaptation

Walk from daylight into a room lit by warm incandescent lamps and, within a minute or two, the white walls look white again — even though the light reaching your eye has shifted dramatically towards orange. The visual system rescales its channels to discount the illuminant. This is chromatic adaptation, and it is why colour appearance models exist: to predict what a measured colour will look like under a given adaptation state, since measurement alone cannot tell you.

Simultaneous contrast

A grey square looks lighter on a dark ground and darker on a light ground. An identical beige reads pink beside green and green beside pink. The visual system judges colour relative to its surround, not absolutely. Any designer who has approved a swatch in isolation and then been surprised by it in situ has met this effect. It is also why colour proofing is done against a controlled neutral grey surround rather than whatever happens to be on the desk.

Colour vision deficiency: what it actually is

The common term is colour blindness, which is misleading. Almost nobody sees the world in greyscale. The overwhelming majority of cases involve one cone type being shifted in sensitivity or absent, which compresses a region of colour space rather than removing colour altogether.

The genes for the L and M photopigments sit on the X chromosome and are adjacent and highly similar, which makes them prone to unequal recombination. Because men have a single X chromosome, an affected copy has no backup. The result is a strong sex asymmetry: roughly eight per cent of men and around 0.5 per cent of women of northern European ancestry have some form of red-green deficiency. Rates vary by population.

  • Deuteranomaly. The most common form, around five per cent of men. The M pigment is shifted towards the L pigment, narrowing the red-green channel. Reds, greens, browns and oranges become harder to separate, particularly at low saturation.
  • Protanomaly and protanopia. The L pigment is shifted or absent. As well as red-green confusion, reds appear notably darker — a red warning light can look dim or nearly black, which has genuine safety implications.
  • Deuteranopia. The M pigment is absent entirely. Red-green discrimination collapses; colours are sorted mainly along a blue-yellow axis plus lightness.
  • Tritan defects. S cone problems, affecting blue-yellow discrimination. Rare and inherited on an autosome, so they affect men and women about equally. Acquired tritan-like loss also appears with ageing, diabetes and some medications.
  • Monochromacy. Genuinely rare, in the order of one in thirty thousand, and usually accompanied by poor acuity and light sensitivity because it involves rod-based or single-cone vision.

What it feels like from the inside

People with congenital deficiency are not aware of missing anything, because they have never had the comparison. What they notice is friction: ripe fruit that is hard to judge, status indicators that all look the same, maps and charts where two categories merge, and a lifetime of being told a shirt is a different colour than they thought. Simulation images are useful for designers but should be treated as approximations rather than a window into someone else’s experience.

Design consequences you can act on

The practical upshot is a small set of habits that cost almost nothing and remove most of the friction.

  • Never encode meaning in hue alone. Add a second channel — shape, position, texture, a label, a line style. A chart legend that says “the red line” fails; a chart where the red line is also dashed does not.
  • Vary lightness, not just hue. Deficiency compresses hue discrimination but leaves the achromatic channel intact. Two colours that differ substantially in lightness stay distinguishable to essentially everyone.
  • Avoid the classic traps. Red-on-green, red-on-brown, green-on-brown, and mid-saturation blue-purple pairs are the reliable failure cases. The old red-amber-green status convention is the single most common offender in software.
  • Test with a simulator. Browser developer tools and most design applications can now simulate deuteranopia, protanopia and tritanopia. Thirty seconds of checking catches most problems.
  • Do not rely on saturation alone for emphasis. Saturated colours are more affected by deficiency than desaturated ones in some axes, and heavily saturated palettes fatigue all viewers.

Where measurement fits in

Colorimetry — the numerical description of colour — is built on a model of the average human observer, derived from experiments in which people matched mixtures of primaries. That model works remarkably well, which is why colour management functions at all. But it describes a standard observer, not any actual person. Real observers vary in lens yellowing with age, in macular pigment density, in the exact peak sensitivities of their pigments, and of course in the presence of deficiency.

This is why two colourists can genuinely disagree about a match that the instrument says is fine, and why critical decisions are made under standardised viewing conditions with more than one pair of eyes. The instrument is a reliable referee for reproducibility. It is not an oracle for appearance. Colour perception is a construction, built from three noisy signals and a great deal of inference, and the more clearly you hold that in mind, the fewer surprises you get when the work leaves the studio.