The eye is not a camera: colour, contrast and the physiology behind display rules

A camera records the light that falls on its sensor. An eye compares, subtracts and rebuilds. The colour a person reports is that construction, not a reading of the wavelength. This first of ten units argues that the display rules which survive the widest range of users rest on that measured physiology. The nine that follow each take one limit of the mind that reads them. The strongest counter is that those measurements were taken on average observers under fixed conditions, so the rule borrows more generality than they carry (Mollon, 1982; Pelli and Bex, 2013). Rules drawn from how marks seem to group together rest on much weaker evidence, because those principles describe an effect rather than fix its size (Ware, 2019). Read a layout judgement as a prediction to test, not a specification to obey.

Colour is built, not received

The retina contains three cone types. Gegenfurtner and Kiper (2003) show that the cones feed three colour-opponent channels, which the cortex combines into the hues that a person perceives.

That construction depends on context. Fairchild (2013) sets out the models that predict how a colour appears under different viewing conditions, including adaptation and the surrounding field. A colour specified in isolation need not survive a real screen. Mollon (1982) surveyed the individual differences in the cone fundamentals, including the anomalous trichromacies in which colour matches differ from normal observers. A hue is a prediction about an observer, and two observers may not match.

Work a case in which a colour is specified in isolation. A designer sets a mid-grey status chip in the design file. The designer checks it against the page background and ships it. On the running dashboard the chip sits inside a dark panel. It is judged lighter than intended and stops reading as secondary. The mechanism is the one that Fairchild models. The eye adapts to the field around the sample, so the surround sets the appearance of a fixed value. The case shows that a specified colour is a prediction. It does not say by how much the prediction will miss.

Why the red-green channel fails for many people

If colour varies between normal observers, it varies far more when a channel is weak or absent. Nathans, Thomas and Hogness (1986) sequenced the genes for the three photopigments. They found that the red and green pigments share about 96 per cent of their sequence. The two genes also sit in a head-to-tail array on the X chromosome. Neitz and Neitz (2011) show that the common red-green deficiencies follow from rearrangements of exactly those adjacent genes.

Because the genes sit on the X chromosome, the commonest deficiencies are far more frequent in men than in women. The deficiency is therefore a design fact rather than a rare exception. A red and a green that differ plainly to one reader can be one colour to another. Design rule: never let colour carry meaning alone.

Work the failure through on a form. A required field is outlined red and a completed field green, so the meaning rides on the channel that varies most. For a reader with a deuteranomaly the two outlines can collapse towards one colour. The field that still needs attention then reads as done. Sajek, Korotenko and Kyrychok (2025) tested schemes of this kind against readers grouped by deficiency type. They found that the best choice depended on the type, so a better red does not repair it. The case shows that the code carries the fault, and not the colour. It does not show which redundant cue a given reader will use.

Contrast, not hue, decides whether text can be read

Legibility turns on a different quantity. Campbell and Robson (1968) measured the visibility of gratings at different spatial frequencies. They read the result as evidence that the visual system contains channels tuned to different frequencies. Contrast, computed separately from hue, is what the reading eye depends on.

Legge, Rubin and Luebker (1987) measured reading rate against contrast and character size. Rates were highest, about 350 words a minute, for letters between about a quarter of a degree and two degrees. Rate fell by less than a factor of two for a tenfold reduction in contrast, and more steeply for very small and very large letters. Readers with low vision need large characters, so they depend on high contrast.

Contrast, then, is closer to a threshold than to a graded scale. Pelli and Bex (2013) cite Fechner's 1860 figure that threshold contrast is about 1 per cent across a remarkably wide range of targets. They argue that any such figure must be reported with the method that produced it, because a threshold depends on the task and on the observer's criterion. Boyce (2014) adds the setting: ambient light is part of the display system.

Work the room as a case. A screen measures a text-to-background contrast ratio of 5 to 1 in a dim office, and there it reads comfortably. Move it beside a window with daylight behind the reader. Ambient light lifts the luminance of the dark areas, so the effective ratio falls. Text that sat on the plateau that Legge measured slides towards his cliff. Nothing about the screen changed. The case shows why Boyce treats ambient light as part of the display system. It does not supply a corrected second threshold.

The standard sets a floor, not a target

Hall and Hanna (2004) had 136 participants read two web pages under one of four colour combinations. Greater contrast generally gave greater readability, but colour did not significantly change retention. Preferred colours drew higher aesthetic ratings and stronger intention to buy.

Vienot, Brettel and Mollon (1999) supply the instrument: colourmaps that show a designer with normal vision how a display appears to a protanope or a deuteranope.

WCAG 2.2 sets a contrast ratio of at least 4.5 to 1 for ordinary text and 3 to 1 for large text, with defined exceptions (W3C, 2024). Those floors are unchanged from version 2.1. Version 2.2 added a minimum target size of 24 by 24 pixels, a measured rule rather than a preference, and the large-text category names reduced acuity, a condition of ordinary later life. Those values keep text legible for a wide range of readers. They are not a target, and passing them does not make a colour scheme good.

Contested question: what does a contrast ratio measure?

The literature disagrees about what a single ratio captures. The standard treats legibility as a property of a luminance ratio between text and background, and it states fixed floors (W3C, 2024). Pelli and Bex (2013) show that a threshold depends on the task, the target and the criterion, so no single ratio can describe what a reader will see. Legge, Rubin and Luebker (1987) found the relation non-uniform, so the same ratio can sit on a plateau for one reader and a cliff for another. The two sides are not equally strong. The floor is a convention about ordinary reading, while the critique rests on thresholds that move with the method.

A second disagreement is whether hue can safely carry meaning. If a channel can be absent, a hue-only code has a failure mode that the designer cannot detect from their own screen (Nathans et al., 1986; Neitz and Neitz, 2011). Vienot, Brettel and Mollon (1999) supplied the means to see that failure. Sajek, Korotenko and Kyrychok (2025) show that the best remedy depends on which deficiency is present, which cuts against a single universal palette. Here the evidence is uneven. The genetics and the deficiency tests are close to decisive, because they name a mechanism and a population. The claim that one palette fits every reader is the weakest, because the studies that would test it use a few schemes and small groups.

Specific evidence would settle both. For the ratio, the test is a reading study that holds contrast constant while varying size, luminance and surround. It measures reading rate rather than detection, which tests whether one number can carry the standard's claim. For the hue question, the test compares a redundant code against a hue-only code across several deficiency types on the same task. That comparison tests whether redundancy removes the failure or merely reduces it. Until then, the floor is defensible and the optimum is not established.

While the dispute is open, a designer can act on what both readings allow. Treat the floor as the minimum that keeps text legible, and not as a target that settles the question. Meet it in the room where the screen will be used. Carry meaning on a second channel wherever hue is doing work, and check the result with a colourmap simulation rather than trusting your own screen (Vienot et al., 1999). Where a claim turns on what one reader will see, test it with that reader.

Where the argument is heading

Sajek, Korotenko and Kyrychok (2025) test schemes against readers with named deficiencies rather than assuming one accessible palette. Pelli and Bex (2013) insist that a threshold be reported with the method that produced it.

The sources this unit rests on

• Boyce, P. R. (2014). Human Factors in Lighting (3rd ed.). CRC Press. doi:10.1201/b16707 A handbook of lighting on how luminance, contrast and adaptation shape visual performance, treating ambient light as part of the display system. It grounds the claim that a screen cannot be judged apart from its room.
• Campbell, F. W. and Robson, J. G. (1968). Application of Fourier analysis to the visibility of gratings. Journal of Physiology, 197(3), 551-566. doi:10.1113/jphysiol.1968.sp008574 Measured the visibility of gratings at different spatial frequencies and read the result as evidence for channels tuned to different frequencies. It is the founding result behind the claim that legibility turns on contrast and spatial frequency rather than hue.
• Fairchild, M. D. (2013). Colour Appearance Models (3rd ed.). Wiley. doi:10.1002/9781118653128 Sets out the models that predict how a colour appears under different viewing conditions, including adaptation and surround effects. It shows that a colour specified in isolation may not survive a real screen, so a specified colour is a prediction to verify.
• Gegenfurtner, K. R. and Kiper, D. C. (2003). Colour vision. Annual Review of Neuroscience, 26, 181-206. doi:10.1146/annurev.neuro.26.041002.131116 Shows that the three cone types feed three colour-opponent channels, which the cortex combines into perceived hues. It treats colour as a construction whose dimensions are not the dimensions of light.
• Hall, R. H. and Hanna, P. (2004). The impact of web page text-background colour combinations on readability, retention, aesthetics and behavioural intention. Behaviour and Information Technology, 23(3), 183-195. doi:10.1080/01449290410001669932 Had 136 participants read two web pages under one of four colour combinations: greater contrast gave greater readability, colour did not significantly change retention, and preferred colours raised aesthetic ratings and purchase intention. The interface-level evidence that colour moves performance and affect, and that the two are not aligned.
• Legge, G. E., Rubin, G. S. and Luebker, A. (1987). Psychophysics of reading V: the role of contrast in normal vision. Vision Research, 27(7), 1165-1177. doi:10.1016/0042-6989(87)90028-9 Measured reading rate against contrast and character size, finding a plateau with rate falling by less than a factor of two for a tenfold contrast reduction, and steeper decline at the extremes. It is the basis for treating contrast as a threshold rather than a graded scale.
• Mollon, J. D. (1982). Colour vision. Annual Review of Psychology, 33, 41-85. doi:10.1146/annurev.ps.33.020182.000353 Sets out the receptor and neural basis of colour, the evidence for opponent processing and the individual differences between observers, including the anomalous trichromacies in which colour matches differ from those of normal observers. It supports the argument that meaning should be encoded redundantly.
• Nathans, J., Thomas, D. and Hogness, D. S. (1986). Molecular genetics of human colour vision: the genes encoding blue, green, and red pigments. Science, 232(4747), 193-202. doi:10.1126/science.2937147 Sequenced the genes for the three photopigments and found the red and green pigments share about 96 per cent of their sequence and sit in a head-to-tail array on the X chromosome. The arrangement explains the existence and the male bias of red-green deficiency.
• Neitz, J. and Neitz, M. (2011). The genetics of normal and defective colour vision. Vision Research, 51(7), 633-651. doi:10.1016/j.visres.2010.12.002 Shows the common red-green deficiencies follow from mutations and rearrangements of the opsin genes on the X chromosome. It supplies the population evidence behind the contrast-based design rule.
• Pelli, D. G. and Bex, P. (2013). Measuring contrast sensitivity. Vision Research, 90, 10-14. doi:10.1016/j.visres.2013.04.015 Reports that threshold contrast for detection is about 1 per cent across a remarkably wide range of targets and conditions, and argues a threshold must be reported with its method because it depends on task and criterion. It is the caution attached to every contrast figure, including the standard's ratio.
• Sajek, D., Korotenko, O. and Kyrychok, T. (2025). Research on the accessibility of different colour schemes for web resources for people with colour vision deficiency. Journal of Imaging, 11(8), 268. doi:10.3390/jimaging11080268 Tested web colour schemes with readers who have different types of colour vision deficiency and found accessibility depended on the scheme and the deficiency type. It turns colour accessibility into a testable design decision rather than a rule of thumb.
• Vienot, F., Brettel, H. and Mollon, J. D. (1999). Digital video colourmaps for checking the legibility of displays by dichromats. Colour Research and Application, 24(4), 243-252. doi:10.1002/(sici)1520-6378(199908)24:4<243::aid-col5>3.0.co;2-3 Provides colourmaps, built from the LMS specification of a standard monitor, that let a designer with normal vision see how a display appears to a protanope or a deuteranope. It makes colour-blindness simulation a practical check rather than a thought experiment.
• Ware, C. (2019). Information Visualization: Perception for Design (4th ed.). Morgan Kaufmann. https://shop.elsevier.com/books/information-visualization/ware/978-0-12-812875-6 Synthesises vision science for display design and offers the colour material as measurement and the gestalt grouping laws as heuristics. It is the source of the claim that the colour rules rest on stronger evidence than the grouping rules.
• W3C. (2024). Web Content Accessibility Guidelines (WCAG) 2.2. W3C Recommendation, 12 December 2024. https://www.w3.org/TR/WCAG22/ Sets contrast-ratio floors of 4.5 to 1 for ordinary text and 3 to 1 for large text, unchanged from version 2.1, and adds a 24 by 24 pixel minimum target size. It is the current Recommendation, which the W3C advises in place of version 2.1, and the standard the unit's rules are measured against.