Color Theory curriculum

Change any one of the three and the color you report can change: the same shirt under another lamp, the same lamp on another shirt, the same scene seen by someone missing a cone type. Asking what color a thing "really" is skips two of the three.

NASA gives the usual human range as about 380 nm at the violet end to about 700 nm at the red end. Radiation outside that slice is still light in the physical sense. Your cones are tuned to this band and do not report the rest.

Frequency is the speed of light divided by wavelength, so 700 nm is about 428 terahertz and 380 nm about 789. The ratio is 1.84. A doubling is one octave in music, so every color in the rainbow fits inside less than one. Young, arguing from his limited receptors and not from this span, concluded in 1801 that colors could not imitate more than a very simple melody.

NASA's examples use stars: at about 3,000°C a star looks reddish, the Sun at about 5,500°C peaks in yellow, and at about 12,000°C a star looks blue. Hold this beside the artist's habit of calling blue a cool color.

Sunlight meets gas and particles in the atmosphere and scatters in all directions, and the shorter blue waves scatter most. No blue material is involved. Color here comes from a size relationship between waves and what they strike.

Newton made the point in 1704 with an analogy to sound: a string trembles, the air carries the motion, and only the listener hears a tone. Likewise a ray has a power to stir a color sensation. Treat a phrase like "red light" as shorthand for "light that usually makes people report red."

Glass bends each wavelength by a slightly different angle, spreading sunlight into a band. Newton's check was to refract one separated color again and watch it stay the same color. The mistake to avoid is thinking the prism adds color to white light.

Newton showed that the rays a prism separates can be brought back together to make white again, and he treated white and the grays as mixtures in due proportion. White names a balance your eye reads as neutral, not a single kind of light.

In Newton's terms the color of a body is a disposition to return certain rays more copiously. A red object absorbs much of the short-wave light and returns more of the long-wave light. Change the light it receives and what it can return changes with it.

Newton put paper, ashes, red lead, grass and blue flowers into a single separated color and found each took on that light's color. They differed only in how bright they looked. White paper even looked brighter than red lead in red light, so "red" paint is not the best reflector of red.

Mixing the two ends of the spectrum, red and violet, gave Newton purples unlike the color of any single separated light. Rainbow order runs from red to violet in a line. The wheel closes that line into a circle only by adding hues the spectrum lacks.

NOAA describes water acting like a filter: red is taken out and what remains looks blue. Floating sediment and particles can push the color toward green or red. That is a different route to blue from the sky's, and you should not explain one with the other.

A star emits, the sky scatters, the sea absorbs. When someone explains a color, first ask which route is at work, because each changes under different conditions: a hotter star, a dustier sky, deeper or muddier water.

Reviews label them L, M and S, for long, medium and short. Reported peaks differ by a few nanometers from source to source (another review gives 560, 531 and 419), so treat them as approximate. These peaks describe sensitivity, not the color you see.

The L cone peaks near 560 nm, about 140 nm short of the 700 nm red end of the visible band. All three cones respond across wide ranges. A long, medium, short label avoids the false picture of one cone per named color.

A receptor's activity depends on both how much light arrives and how well its wavelength suits the pigment. A dim favored wavelength and a bright unfavored one can produce the same response. So an animal with one class of photoreceptor cannot perceive color differences.

Cone sensitivities overlap, so most wavelengths drive more than one type. The brain receives signals it can compare, and a different balance across cones means a different color. At least two spectrally distinct receptor classes are needed for any color vision at all.

Rods respond in very dark conditions and stop contributing in normal light. They process achromatic information, which is why a moonlit garden loses its hues while you can still see shapes. If a color question assumes night vision, the answer usually changes.

Young proposed it, and Maxwell and Helmholtz extended it: three receptor kinds, with mixtures of three lights matching most colors. It accounts for mixing very well. It leaves negative afterimages unexplained, which is the gap the opponent account fills.

In an 1801 lecture he argued that each point of the retina could not hold particles tuned to every possible wave, so the number must be limited. Three sufficed, with a wave in between moving two kinds at once. Note how far an argument about a limit carried, with no anatomy.

His 1801 lecture used red, yellow and blue, and treated green light as stirring the yellow and blue particles equally. Later work moved to red, green and violet. This is a useful reminder that the idea of three came first and the choice of which three was revised.

A 2021 review places the long and medium opsin genes together on the X and the short one elsewhere. That single fact predicts a great deal: red-green defects track sex, while blue defects do not follow the same pattern. Remember the location before you memorize the prevalence.

A mean of about 2.7 L cones per M cone (about 73% L) has been reported in color-normal American men of European ancestry, and the mean differs between groups. Individuals vary around it. Do not picture the retina as equal thirds of three cone types.

Early vertebrates already carried four cone opsin families, and placental mammals kept only two, so they are dichromats. Primate trichromacy arose from that dichromatic ancestor. Human three-cone vision is not the upper limit or the default in the animal world.

In Old World primates a duplication placed two L/M genes on one X chromosome, so both sexes are trichromatic. Most New World monkeys vary one gene between versions, so one group can hold both trichromats and dichromats, a ready-made comparison for field researchers.

The National Eye Institute notes small differences between people even when no deficiency is present, and it notes that age can degrade color vision, often through cataracts. A "normal observer" is a statistical convenience rather than a single shared experience.

His idea was a pull between pairs rather than three separate receptors' reports. Later work by Hurvich and Jameson showed that one color's sensation can be counterbalanced by its opposite. It is the origin of the opposing axes you meet again in color spaces like Lab.

Three cone types capture light, and later cells compare what the cones report. Learners often ask which theory won. Current accounts use both: the first explains matching, the second explains afterimages and how signals are recoded on the way to the brain.

These comparison cells in the lateral geniculate nucleus, a relay between retina and cortex, are the physiological basis for chromatic opponency. A cell like this signals the difference between inputs, so its output says something no single cone can.

L and M cones respond to much of the same light, so their raw signals repeat each other. Taking differences keeps what is new and reduces repetition in natural images. This is an efficiency argument, and it explains why comparison would be worth wiring at all.

These cells respond to color edges rather than to a uniform patch. That spatial comparison is one reason a color's appearance depends on its neighbors, and it links this chapter to simultaneous contrast and constancy later on.

Contrast sensitivity for a red-green pattern at equal luminance is low-pass, while for brightness patterns it is band-pass. That supports treating color and luminance as partly separate channels. In practice, thin lines and small text are safer with a lightness difference.

A 2022 macaque imaging study found organized circuits in primary visual cortex that combine L/M and S/(L+M) signals along lines predicted from studies of perception, with hue preferences arranged in pinwheels. So the step from cone comparisons to perceived hue begins later, in the cortex.

The idea that red, yellow, green and blue look unmixed goes back at least to the 17th century, yet solid neural evidence has been hard to find. A 2017 study reported an earlier brain response at about 230 ms for those hues, at a post-perceptual stage. That is one marker, not a settled explanation.

A 2011 study linked that boundary to the sign of the L-minus-M cone contrast, and found the same two-way split in macaque visual cortex. It is one proposed biological basis for a constraint on naming, and it does not explain how each language then divides its colors.

A match is decided at the cones. Two stimuli can differ across the whole spectrum and still look identical to an observer in the same setting. Keep two words apart: a stimulus is its spectrum, and a color is what the visual system makes of it.

A display does not reproduce the spectrum of a lemon. It emits three primaries in amounts that produce about the same cone response, and that is enough for a match. Metamerism is the reason three primaries can stand in for the world's unlimited spectra.

That observer is a set of color-matching functions, a statistical stand-in for human vision. A display tuned to it is "accurate" for the stand-in. The standard does not exactly represent any given person, which sets up the disagreement described next.

It is the name for two stimuli with different spectra that look the same to one observer and different to another. Studies describe this variability as significant. When two people with normal color vision disagree about a match, neither needs to be wrong.

Laser cinema projectors reach more colors because each primary is very narrow. The same narrowness means small personal differences in cone sensitivity shift the match more, so a 2023 index computed from cone data predicts more disagreement than with broader lamps. Its authors still call for direct tests with observers.

Two samples with different reflectance spectra may match in the store and differ at home, because a change of light changes what each sends to the cones. If a match matters, for a repair or a replacement part, compare the two under the light they will live in.

A 2019 review gives lights at 467 and 582 nm, 115 nm apart, that produce the same photon catch in M cones and so the same M-cone response. Any cone sends only one number. Spectral detail is lost at this first step, and later stages cannot recover it.

Munsell noted in 1907 that spectral red and green reach the eye as yellow, to the surprise of people who knew only paints. A single wavelength near yellow and a mixture with none at that wavelength can match. So "this light is yellow" says nothing about which wavelengths are present.

If two surfaces match under one light but not under another, the light reaching your eye does not fix which surface you are looking at. That is why the argument about how good constancy really is continues, and it returns in the chapter on context.

Two beams on one wall deliver both beams' light to your eye. Nothing is taken away, so the mixture is at least as bright as either part. If a mixing question involves a screen, a spotlight or a projector, this is the rule to use, and paint intuition will mislead you.

Munsell described sending red and green through slits onto one screen and seeing yellow, with neither red nor green visible. On paper this looks impossible. Red and green pigments give a yellow-gray at best, because paint removes light and these beams are adding it.

Spectral yellow and blue unite in white, as Munsell pointed out to painters of his day. A yellow beam already stimulates L and M cones, and the blue adds S, so all three are engaged. It is the cleanest single case for showing why light and paint need separate rules.

Maxwell's discs, used in Munsell's teaching, carry adjustable sectors that blur into one color when spun quickly. No light is combined on the surface; the eye merges what it receives. Munsell used them to test claimed complements by whether the pair could reach a neutral gray.

Newton noted that froth shows many colors up close, while from far enough away that you cannot separate them, it looks white. Halftone dots lean on the same limit. This is mixing by the eye rather than by combining pigments in a pot.

The CSS specification's example: two identical lights in one fixture measure exactly double one light. RGB code values are not like this, because they are gamma encoded. Add or average encoded numbers and you get a result that is not what the lights would give.

Decoding with the sRGB curve gives 0.216 for 128 out of 255. To get half the light you need a code value near 188. So #808080 is not half as much light as white, however it looks in a slider. Run the formula when anything physical depends on it.

The midpoint of pure red and pure green is rgb(128 128 0) if you average code values, with a luminance of about 0.20. Averaging the light gives a luminance of 0.46, which is about rgb(188 188 0). The specification warns that sRGB mixing produces overly dark or grayish results.

Plotted on a chromaticity diagram, three primaries form a triangle and every mixture lands inside it. Colors beyond it are out of gamut for that device. Points beyond the outer curve of pure wavelengths are imaginary colors, which no light could produce at all.

Beams add, so they head to white. Stirred pigments mostly lose light, so they head to dark and dull. Dots too small to resolve blend in the eye, with the result lying between their colors. When someone says "mixing colors," ask which of the three they mean.

If one filter passes 80% of a wavelength and the next passes 50%, then 40% survives, since 0.8 times 0.5 is 0.4. Any wavelength that either filter blocks completely stays blocked. Ignoring surface reflections, work each wavelength separately and you can predict what a stack lets through.

In the simplest model, cyan removes red, magenta removes green and yellow removes blue. Cyan plus yellow removes red and blue, which leaves green. This idealized bookkeeping explains why those three are the printing set. Real inks depart from it, which is why accurate conversion needs a device profile.

Lovibond contrasted two ways of isolating a color: spreading it out by refraction, or absorbing everything else. In the second, what is left is all you get, so each new absorber can only darken the result. That is the core reason mixing dyes moves toward black.

Lovibond's color measuring scale built units of tinted glass so that a unit of red, a unit of yellow and a unit of blue passed a neutral gray. The units had to be made equivalent first. Three random colored filters will not reach neutral on their own, since a small change in any one upsets the balance.

Lovibond argued that a reflected or transmitted color follows the makeup of the light it receives, so measuring color calls for a colorless source. In 1914 that meant diffused daylight. A dye that looks one way at noon can look different under a lamp.

Newton tried to mix painters' powders into white and got gray or a dun brown. His reason still holds: colored powders suppress a large share of the light they receive, and do not return even their own color as well as white paper does. Stirring only adds more loss.

Munsell blamed the absorbing action of pigments for the belief that green is a mixture of yellow and blue. The paints together remove most of the spectrum and leave what both reflect. Green here is a leftover from subtraction, not a sum of two sensations.

Munsell's list of pairs reads like this: red and green make a yellow-gray, yellow and blue a green-gray, green and purple a blue-gray. Each mixture is a weak-chroma color. Learners often expect a clean mid-hue, then blame the paint when a gray appears.

Lovibond's 1914 survey listed theories with seven, six, five, four and three primaries, from Newton's later seven to Young and Helmholtz's three, each advanced on experimental grounds of its day. No set has a special claim outside a stated use. Ask "primary for what?" before you accept any list.

In Lovibond's account, scientists took red, green and violet from mixing colored lights, and artists took red, yellow and blue from mixing pigments. Both were reporting real experience. Most arguments over "the" primaries come from assuming one kind of mixing covers both.

He wrote of the mischief of that teaching, noting that Rood had challenged it in 1879, and that Brewster's scheme was dropped once color vision was traced to red, green and violet-blue. Red, yellow and blue remain a workable painter's convention, without being fundamental.

His circle spaced red, yellow, green, blue and purple evenly, with five intermediates between them. His complaint against spacing red, yellow and blue evenly was that it spreads the warm field too far and shrinks the blue one, so balance fails. A different starting set gives a different geometry, which is exactly why a primary set is a design decision.

Field's Chromatography lists orange, green and purple as secondaries and citrine, russet and olive as tertiaries, each resolved back into red, yellow and blue. The ladder classifies pigments, and it says nothing about how the eye works. Use it as a record of painters' practice.

sRGB, Display P3 and Rec. 2020 all have red, green and blue primaries, placed differently. Maximum red in Rec. 2020 is not the same color as maximum red in sRGB. This is why the specification requires a tagged image to be converted, not reused as raw numbers.

In Opticks he split the circumference into seven parts in proportion to the tones of an octave, for red, orange, yellow, green, blue, indigo and violet. He reported that his measured spectrum fell into those divisions. Young and Munsell both later doubted that color could be handled like music.

He placed each ingredient as a weight at the center of gravity of its arc, then read the mixture from the balance point. The direction from the center names the hue, and the distance from the center gives its fullness, or how far it is from white. Mixtures in modern chromaticity diagrams fall between their components in a similar way.

A 2021 review credits Newton with adding orange and indigo to the rainbow to reach seven. The spectrum is continuous, and where you cut it depends on the words you have. The same source notes that language researchers count eleven basic color categories, again a different number.

His opposite pairs were yellow and purple, orange and blue, red and green, because staring at one seemed to evoke the other. That is a claim about perception rather than about light. It explains why his pairing does not match the pairs you get from mixing beams.

In "Theory of Colours" he explained the prism on principles different from Newton's and called color a degree of darkness. Even his 1840 translator noted the opposition this met. Physics followed Newton. What lasted is his careful record of afterimages and shadows.

Goethe's setup: place a pencil on white paper at twilight with a candle beside it, so that weak daylight lights the candle's shadow. He reported that the shadow looks vivid blue, and credited the eye's response to the candle-tinted page. That shadow is lit by daylight alone, so the setup cannot rule out the color of the daylight itself.

Researchers note that the word has mutually distinct senses. By light mixture, red pairs with cyan in the RGB scheme. In opponent terms and on the painter's wheel, red pairs with green. Munsell paired red with blue-green. Before you use the word, say which definition you mean.

A 2020 study concluded that a circle cannot place afterimage hue pairs as true diameters, whether it follows RGB pairs or Hering's. Curved arcs can rescue the RGB scheme, while Hering's fails even then. The neat wheel is still handy for artists; just do not treat it as a measurement of the eye.

HSL puts red at 0°, yellow at 60°, green at 120°, in even sixties. Running those colors through the Oklab formula gives hue angles near 29, 110, 142, 195, 264 and 328 degrees. The gap from yellow to green is 33°, and from red to yellow 81°. An even wheel is a convenience of the model.

Red (255, 0, 0) inverts to cyan (0, 255, 255), and the two add to white. In HSL terms that is a 180° rotation, so blue at 240° pairs with yellow at 60°. It is a fair definition for light on one screen, and it does not predict what paint will do.

Munsell opened with Stevenson writing from Samoa, trying to order wallpaper by letter with words like "topazy yellow." Names borrowed from gems and plants vary with the speaker. His whole system began as a response to that communication failure.

Hue says which color family, value says how light, and chroma says how far from gray. Changing one need not change the others. Munsell's contribution was making each a measured scale, so "a bit brighter" can be rewritten as a change in value or a change in chroma.

R, Y, G, B and P are the principals; YR, GY, BG, PB and RP sit between them. A color can then be given without a poetic name. Note where orange went: it is only the intermediate YR, while purple is a principal. His stated concern is balance around the circle, so any reason for each single hue is an inference.

The grays are written N1, N2 and so on, with middle gray at 5. Every hue is then placed against the same ladder. So you can compare the lightness of a red and a green directly, even when their hues make that hard to judge by eye.

He found school gray scales varying by over 10 per cent, and one supposed middle gray returned just 42% of the light. His remedy was an instrument with a dial. A scale that runs through the same eyes it is correcting cannot find its own error.

Munsell's dial treated 42% as nearly a whole step below middle, which implies middle near 50% of the light. On the modern Lab scale, the visual midpoint L = 50 is about 18% luminance, by the published formula. Always ask whether "middle" means energy or appearance.

Pigments are very unequal in strength, Munsell found, so a strict sphere does not fit; he called for a color tree with branches of different lengths. The top of a chroma scale can move outward when a stronger pigment appears. Value is bounded in a way chroma is not.

Hue comes first, value sits above the line and chroma below it. Munsell gave R 5/9 as the pigment vermilion. Read R 5/1, R 5/5, R 5/9 as one hue at one value going from grayish to strong. The written form lets you specify a change exactly.

Munsell reported that many children found it hard to state more than five chroma steps, though they readily made ten value steps and twenty to thirty-five hue steps. This is one teacher's classroom observation and not a controlled measure, but it explains why his tree is shaped as it is.

A line through the center joins opposite qualities, so R 5/5 balances BG 5/5. His course also pairs a light weak red, R 7/3, with a dark strong blue-green, BG 3/7, and uses area to help balance unequal pairs. A complement here depends on all three dimensions, not on hue alone.

Munsell's case: the violet looks like the buttercup's opposite, yet at chroma 4 it falls five steps short of the buttercup's 9, so the pair does not balance in gray. No blue flower, he wrote, could match that chroma. This separates the "opposite on the wheel" idea from a measured balance.

Munsell objected to writers who said one yellow "shades into green" or spoke of "a brighter shade." One of those is a hue change and the other a chroma change. Using the three axes strictly is what makes a color instruction checkable.

The World Color Survey collected names for 330 chips from 110 languages of nonindustrialized societies, and chip catalogs serve in constancy tests too. Designed for near-equal perceptual spacing, they gave researchers a shared ruler, which is what Munsell had wanted.

A triple like 0.5 0.5 0.1 could be many colors. The specification insists that an image tagged Rec. 2020 shown on a Display P3 screen be converted, because treating the values as P3 values gives wrong colors. Ask for the space first and the numbers second.

HTML colors and untagged images are assumed to be sRGB unless otherwise specified. Video is different: its default depends on resolution, from BT.601 below 720p, through BT.709 at 1080p, to BT.2020 at 4K and above. A mismatch can appear with no error anywhere.

The red primary sits at chromaticity x 0.640, y 0.330; green at 0.300, 0.600; blue at 0.150, 0.060; and the white is D65. Together with the encoding curve those settle what any code value means. Change one and you have a different space.

D50 approximates sunlight on a clear day at a color temperature of 5000 K. It is the white used for Lab, and measuring instruments use it when they report Lab. D65 is the white of most RGB spaces. Converting between them is a modeled adjustment, usually the linear Bradford transform.

In HSL, sRGB blue and sRGB yellow both have a lightness of 50%. In Lab they sit at about 30 and 98. HSL was built as a friendlier rearrangement of RGB and carries none of the eye's unequal response to the primaries. Never use it to judge readability.

Full-saturation yellow and full-saturation blue both read 100% in HSL. Their OkLCh chroma values, computed with the published matrices, are about 0.21 and 0.31. Two colors reading identically on a slider can differ a lot in how far they stand from gray.

L runs from 0 for black to 100 for diffuse white, and L = 50 is designed to look like mid gray. Equal L steps look about equal while equal energy steps do not. It is the usual form for reporting physical color measurements and for moving between devices.

Positive a is a purplish red and negative a a green; positive b is yellow and negative b a blue-violet. You can see Hering's opposed pairs in that layout. A color near the L axis has small a and b and looks gray. Move outward and it gains chroma.

The specification lists known faults: in the region of LCH hue 270° to 330°, a saturated blue reduced in chroma turns visibly purple while its hue angle stays the same. Gradients and gamut mapping suffer. Knowing the fault matters more than memorizing the region.

Plain distance in Lab overstates differences among vivid colors, so later formulas added corrections. Delta E 2000 is the current industry standard. It helps with visual difference but does not fix the hue bending in blues. Report which formula you used along with any number.

The specification credits it with better hue linearity, hue uniformity and chroma uniformity than LCH. Blues hold their hue as chroma drops. It uses D65 as its white, and like Lab it is not a final model of vision, so treat it as the better tool for now.

Because the space is more even, the specification's color difference there is the root sum of squares across L, a and b. Lab needs a complicated formula to do the same job. That simplicity is a large part of its appeal for software.

CIE LCH chroma can reach 200 or more, while OkLCh chroma runs to about 0.5. Hue angles between the two are similar but not identical. So a value copied from one notation into the other can be nonsense, even though both are labeled L, C and H.

The specification's example is a vivid lime in gamut for Rec. 2020 with values near 0.42 0.98 0.01. In Display P3 the same color needs a negative red and a green above 1. It exists, and that screen cannot show it. So gamut questions always have the form "in gamut for what?"

A color outside the gamut is legal but cannot be shown on this device, so it gets mapped. An imaginary color lies outside the visible range and can never be physically produced. An invalid color, such as one with a misspelled color space name, renders as opaque black under the CSS rule.

The specification's example: linear values 0.5 1 3 clipped to 0.5 1 1 move from an OkLCh hue of 265.1° to 196.1°, a shift of 69°. Blue was three times green and now merely equals it. Clipping can be acceptable when a color is only slightly outside, and it can fail badly otherwise.

Rather than chop each channel, a mapper reduces chroma in a perceptual polar space, because hue changes are particularly objectionable. The usual method holds hue and lightness fixed. It applies only when the destination is a real display or a printer. In-between calculations keep out-of-range values, so precision is not thrown away too soon.

Linear spaces model physically combined lights. Oklab gives perceptually even steps. OkLCh keeps chroma up and avoids a gray middle. sRGB is kept for compatibility despite darker, grayer mixes. CSS now uses Oklab when a feature names no other. Pick by what the mix is meant to model.

The CSS Color 5 draft says device CMYK has no colorimetric interpretation, and that converting it depends on the output device. The numbers are amounts of ink. Ask which inks, on which paper and with what total coverage, because each of those changes the color.

The draft works one case: a value printed for coated SWOP and interpreted through a FOGRA coated profile differs by delta E 2000 of 8.17, which it calls clearly visible. A naive formula with no profile gives 14.3, which it calls very visible. So a conversion with the wrong model can do real harm.

Orange, green and violet inks extend what plain cyan, magenta, yellow and black can reach, and need a matching profile. This is the print counterpart of a wider-gamut display. Adding a primary widens the reachable set, which shows again that the first four were never the limit.

D50 is the white used by ICC profile connections, image editors offering Lab and many measuring devices, while most RGB display spaces use D65. A file crossing between them has to go through chromatic adaptation. When a proof and a screen seem to disagree, the white is a good first question.

The specification treats a screen and a printer alike as destinations needing in-gamut values. A vivid display color can land well inside a printer's gamut, or well outside it. Plan for the smaller gamut when something must look the same on both.

That is why a print's color depends directly on the room's light. Lovibond's concern with a colorless reference light for judging samples comes from the same fact. Judge a print under the light where it will be seen, and do not assume a monitor will agree with it in a dim room.

Each point on the paper either has ink or does not, so the appearance of intermediate tones comes from patterns of dots too fine to resolve individually. This puts the mixing in the eye, the same limit Newton saw in distant froth, and the dots still subtract light from the paper.

Laser projectors have a wider gamut thanks to very narrow primaries, and that same narrowness increases observer metamerism. A theater-goer beside you may not see quite the color you do. Each medium's strength can carry a cost you will not find on the specification sheet.

Each is a rule about angles on a circle, and the circle differs by tradition. A "complementary" pair on a painter's wheel is not the same pair on an RGB wheel. So the scheme cannot be a law of perception by itself, since its output depends on a prior arbitrary choice.

A wheel rule can quickly give you options, and it cannot certify them. Check the lightness difference, the chroma and the area of each color, then look at the real arrangement. State schemes as whose convention they are, and expect them to fail on some pairs.

A 2015 paper traces the belief in harmonious relations between opposites on a wheel to Goethe's 1810 book. Tradition has kept repeating it. Age is not evidence, so treat the belief as a hypothesis that later measurement could support or undercut.

As that paper reports from Schloss and Palmer's work, researchers separate pair preference, pair harmony and figural preference. Earlier studies blurred them. If a claim says color pairs "work," ask which of the three judgments was actually measured.

In the Schloss and Palmer findings summarized there, both pair preference and harmony ratings correlate negatively with hue contrast. Traditional theory expected the reverse. The traditional claim seems to fit figural preference better. You are reading a secondary report, so hold it with care.

The same summary reports that pair preference goes up with larger contrast in lightness, and when the component colors are themselves liked. That may be a better first move than rotating a hue. It also ties preference, which that paper keeps separate from harmony, back to value, the dimension wheels leave out.

That 2015 study linked harmony ratings to activity in the medial orbitofrontal cortex and amygdala, and found ratings for disharmonious pairs less variable between people than ratings for harmonious ones. It is interesting, and 18 volunteers cannot settle a design rule.

His example: a rule of three parts yellow, five red and eight blue fails because it never says which yellow, red or blue. The same words meant different things to each user. Any harmony rule that does not specify value and chroma inherits the same flaw.

Music has measured intervals, while color does not offer a matching set. Munsell judged the two too far apart in their scales for any practical relationship. A chord-like claim about color therefore borrows prestige from a system whose structure it does not share.

From a given green at G 6/5 you can change value along a vertical path, hue along a lateral path, or go inward through neutral to the opposite field. He offered these as paths to try, with taste still choosing. It is a vocabulary for search rather than a proof of beauty.

His scheme had six kinds, three built from closeness in scale or hue or a shared tint of light, and three built from strong differences. Note the source: a dye chemist's long practical observation for tapestry work, available in an 1861 English edition. It describes a craft rather than a tested effect.

In a 2022 experiment, observers chose the original over rotated versions, even for unfamiliar works and even when images were scrambled into small squares. So some preference tracks color distribution as well as subject matter. That says little about any one wheel-based rule.

As director of dyeing at the Gobelins, Chevreul heard that blacks in shadows of blue and violet draperies lacked vigor. He got black wool from other dye works and found them no better. The trouble lay in the colors next to the black, not in the dye. Check the surround before blaming the sample.

When two colors are seen together, each seems pushed away from the other in both hue and depth of tone. The rule predicts direction, not amount. Note what it does not say: nothing in the materials changed. Identical gray patches can look different on different grounds.

Chevreul named these contrast of tone and contrast of color. Beside a light color a dark one looks darker, and a hue shifts away from its neighbor. Either can occur alone. When a match looks wrong, ask which of the two is changing, because the fixes differ.

Chevreul described each of two adjacent colors as if it received the complement of the other. It is a compact way to predict direction: gray beside red drifts toward a greenish look. Treat this as a description of the appearance, not a claim that any light was added.

Chevreul called this mixed contrast. Close one eye, fatigue the other on color A, then look at color B: you see B altered by A's complement. Switch to the rested eye and he reports that B may look shifted the other way, toward A. This separates a change in the eye from a change in the paper.

Explanations have run from pigment bleaching in cones up to adaptation in cortex. Recordings reported in 2012 found primate ganglion cells adapting subtractively and rebounding when a stimulus ends, on time scales slower than expected from receptors alone. Cortex may still modify the signal.

Induction by a surround was stronger for brief presentations of 10 to 40 ms than for long ones of 80 to 640 ms, in earlier work. Reanalysis of 12 small data sets pointed to two factors in each case. The samples are small, so read it as a consistent pattern and not as a settled mechanism.

For brightness, wider stimuli produced weaker simultaneous contrast, and a 2013 study measured the relationship as logarithmic. A center-surround filter matched it closely. So the same two tones can behave differently as a large field and as a thin bar.

Under contrast, a gray patch drifts toward the opposite of what borders it. Under assimilation it drifts toward that border's own color. Learners who know only contrast predict the wrong direction in such cases. Earlier depth studies disagreed on whether contrast survives; a 2020 experiment found it persists, much reduced.

A point can look more like its near neighbors and less like remote ones. A 2019 neural field model treats these as two faces of one spatial process, with a larger measured shift when a pattern is present. So asking "which effect is this?" may have no single answer.

A 2022 study found strong shifts for a center line flanked by white contours on chromatic backgrounds, with different best widths for L/M and for S-cone backgrounds, near 1 minute of arc for the first and about 8 to 17 for the second. Thin colored marks should not be judged from a large sample of the same paint.

The Bezold-Brücke effect is a hue shift with changing luminance, and it is the oldest of these recorded effects. So "the same wavelength at higher brightness" is not guaranteed to be the same hue. This is one more place where a single physical number does not fix the appearance.

The Abney effect is a shift in hue as a sample's purity changes. Adding white light can therefore do more than make a color paler. Keep this beside the blue-to-purple fault in Lab, since a space that tracks hue well has to deal with effects like this one.

The ratio of brightness to luminance varies with wavelength, lowest for yellow near 575 nm and highest toward the spectrum's ends. This is called the Helmholtz-Kohlrausch effect. So two samples with equal measured luminance need not look equally bright, which limits any luminance-only rule.

Your eye receives light that blends illumination with surface reflectance, and the two cannot be separated from that signal alone. A problem like this has no unique solution without assumptions. Your visual system solves it well enough that you rarely notice the change of light.

Lab tasks have reported constancy from 15% to 80%. A 2024 study asked 16 people to match remembered familiar objects under five office lights. They mostly selected the same Munsell chip, and the authors called the result, under ideal conditions, as good as visual memory permits. Both results can hold, because the cues differ.

A 2014 study painted matching 3-D blocks and viewed one set under uniform light and one under directional light. The same paint showed shifts in lightness, hue and chroma. Flat-pattern generalizations did not carry over. Appearance tracks spatial information, not reflectance alone.

In a 2017 study of natural scenes under simulated daylight changes, seven dichromats and four normal trichromats gave no significantly different responses when asked whether a change in the scene was only a change of intensity. The sample is small. Losing a cone class need not mean losing the ability to discount the light.

People saw blue and black, as the dress actually was, or white and gold. The leading account is differing assumptions about the light. A small study of 39 people also reported group differences in macular pigment and visual brain response timing. One small study does not settle those added causes.

The National Eye Institute says most people with it see colors differently and find some hard to tell apart, and that a complete loss is rare. Most manage everyday activities. "Color blind" is the common phrase, and for most people it overstates the loss.

The risk is far greater for men than for women, and mild cases often go unnoticed. Whenever you see a prevalence figure, ask which population it describes, since the number moves between groups, as you will see later in this chapter.

Deuteranomaly is the most common, and shifts certain greens toward red. Protanomaly shifts certain reds toward green and dims them. Both are mild. In protanopia and deuteranopia, a person cannot tell red from green at all. Keep the "-anomaly" (altered) and "-anopia" (missing) endings separate.

Tritanomaly makes blue and green hard to tell apart, as well as yellow and red. Tritanopia adds pairs such as yellow with pink and purple with red, and it also dims colors. The label "blue-yellow" does not mean blue and yellow alone are confused.

People with it see no colors at all, and depending on the type may also see less clearly and be more sensitive to light. One review estimates total color blindness at about 1 in 30,000. Most people called color blind are in a different category entirely.

Males have one X, so one altered copy is enough. Females have two, and both must carry the gene, which is rare. As a rough check, 8% squared is 0.64%, close to the 0.5% commonly reported for women of the same ancestry. This is the arithmetic of X-linked inheritance.

Inherited deficiency stays stable over life. Acquired forms can result from conditions like glaucoma or macular degeneration, from injury to the retina, optic nerve or brain, or from some medicines. A change in color vision in adulthood is a reason to see an eye doctor. This is general information, not medical advice.

The shape stands out for people with typical color vision and blends in for many with a deficiency. Other tests include the anomaloscope, which asks you to match two lights with adjustable knobs, and arranging colored blocks in order. A diagnosis comes from an eye doctor's test.

A 2025 review gives rates of up to those levels for that population and reports lower rates in Asian and African populations. Quoting 8% as a world figure overreaches, though some sources do so. Designing for deficiency is still warranted in every population, since no group is at zero.

Because screening is limited, diagnosis often waits until school age or beyond, and some discover it in adulthood. This has consequences for education and work. So you cannot rely on your audience to tell you they need another cue, because they may not know it themselves.

Across 16 cross-sectional studies it reported 2.71% overall, with 2.13% in males and 0.34% in females, and heterogeneity was high. Those figures come from particular school and community samples. They show why a single headline rate cannot describe the whole world.

He confused scarlet with green and thought a blue tint in his eye fluid filtered long waves. Examination after his death found the fluid clear. Much later, DNA from his preserved eye tissue showed that the middle-wave pigment was the one he lacked. It is a fine example of a testable wrong hypothesis.

Filters try to separate overlapping cone signals. A 2022 study of 16 people with red-green deficiency found no one reclassified on a clinical test while wearing them, and its title says they cannot compensate. A UC Davis study publicized by the National Eye Institute reported benefits after two weeks. Treat them at most as a contrast aid, not a cure.

The same paper lists aviation, rail, marine transport, police, armed forces, fire services and electrical work among fields where a diagnosis may block or impede entry. Rules vary by employer and country. If this affects you, check the specific standard rather than a general list.

A 2014 study found greater variability in naming, yet protans and deutans named colors better than expected, which suggests use of non-chromatic cues. So correct naming does not rule out a deficiency, and a person's correct use of color words tells you little about what they see.

In a 2021 study of 130 men, associations such as red with love and anger, and yellow with joy, were chosen by at least half of both colorblind and non-colorblind participants. That points to learned concepts rather than immediate sensation. Much of color meaning travels with words.

First decode each 8-bit channel with the sRGB curve, then compute 0.2126 R + 0.7152 G + 0.0722 B. The result runs from 0 for black to 1 for white. Skipping the decoding step is a common hand-calculation error, and it gives wrong answers for every mid-tone.

The coefficients give pure sRGB green 0.7152 and pure blue 0.0722, a ratio near 9.9, with red at 0.2126 between them. That is why full-strength colors at the same HSL lightness read so differently. Pure blue text on black is hard to read for a reason you can compute.

Lighter first, so the result is always at least 1. The largest value is white on black: (1 + 0.05) divided by (0 + 0.05) is 21, written 21:1. Forgetting the 0.05 makes any ratio against pure black go to infinity, which is a quick sign of a wrong formula.

Large means at least 18 point, or 14 point bold. Level AAA raises these to 7:1 and 4.5:1. Both are written as "at least", so a pair at 4.48:1 falls short of the AA text threshold even though it looks the same as one that passes.

Computed, they give 4.54:1 and 4.48:1, a single code step apart on each channel. For large text the same flip happens between #949494 at 3.03:1 and #959595 at 2.995:1. Run the calculation, because no one can see this boundary by eye.

Pure red scores 4.00:1 on white and 5.25:1 on black, so it fails text use on white. Pure blue scores 8.59:1 on white and 2.44:1 on black. Pure yellow scores 1.07:1 on white and 19.56:1 on black. Pure green scores 1.37:1 on white.

Pure red on #008000 green is a textbook complementary pair on a painter's wheel, and it computes to only 1.28:1. Their luminances are close. To a reader with a red-green deficiency, this pair may have little left to separate them, so hue difference alone is not enough.

The worst case is a background with luminance near 0.179, where the two ratios meet at the square root of 21, about 4.58:1. Above that luminance, black text wins, and below it white wins. You can compute this from the formula, and it is the idea behind automatic text color.

Gray at code value 118 gives 4.54:1 on white and 4.62:1 on black, so the flip sits near there rather than near 128. That is because luminance is not linear in code values. A guess near the middle of the code range lands too high, and then white text gets chosen where black would score higher.

Under Success Criterion 1.4.11, the visual information needed to identify a control, or the parts of a graphic needed to understand it, must reach 3:1 against adjacent colors. Inactive controls are exempt. A thin chart line can fail this even when the text beside it passes.

WCAG exempts brand names and logos, inactive controls, pure decoration, and words inside a picture whose main content is something else. Exemption is not a guarantee that people can read it. Use this only for elements that really fall into those classes.

Success Criterion 1.4.1, at Level A, covers meaning, actions, prompts and telling elements apart. A form that marks errors only in red fails, whatever its contrast ratio. Pair the color with an icon, text or a shape, and the message survives a monochrome print as well.

Before May 2021, WCAG's definition used 0.03928 in place of 0.04045 as the sRGB linear threshold. WCAG itself notes that the change has no practical effect on these calculations. If two tools differ in the fourth decimal, this is not the reason a pair fails.

The CSS Color 5 draft says it plainly, naming the font, the text size and surrounding colors as further factors. WCAG also warns about very thin strokes. Measured luminance also misses brightness effects like Helmholtz-Kohlrausch. Treat the ratio as a floor, then read the result.

Of 66,253 images from one journal, 85.6% held at least one pair that could trouble deuteranopes. Yet in a sample of 4,964 reviewed by hand, 12.8% were judged hard to interpret, since labels and spacing often rescued them. Simulation helps you screen; it does not settle the question.

A 2020 paper in Nature Communications argues that a data ramp should change by equal perceived amounts per step. Rainbow palettes do not: they draw false edges in parts of the range and bury small changes in others. In effect the map changes your reading of the numbers. Use a perceptually uniform map instead.

Traffic-light red and green at equal lightness can lose the difference altogether for red-green deficient readers. Vary lightness along the scale so the order survives without hue. If you print your map in gray and it still reads in order, it is probably safe in this regard.

Of 711 people in four countries, a classifier guessed which of 12 color terms was rated at 38.7%, against 8.33% by chance. That is well above chance and far from certain. Red was linked decades ago with excitement, joy, hostility and power at once. A one-to-one color chart is not supportable.

That classifier named the country of origin with 80.2% accuracy, against 25% by chance, while the in-group advantage in reading color terms was only 6.1%. So there is a large shared pattern with distinct national differences on top. Both "universal" and "purely cultural" overstate the case.

A 2021 paper reports a large cross-national survey with an average correlation of about 0.88 between countries, for associations given to color terms. Agreement like this concerns words such as "red" more than any measured stimulus. It also says nothing about whether a color changes anyone's feelings or behavior.

A study of 873 Arabic speakers found correlations above 0.90 between term and patch conditions, in lab and online settings, with red and love the top link at 61.5%. That suggests the link lives largely in concepts, not in the immediate visual signal.

Surveys record what people pair with a color. Claims about behavior need experiments, and a 2015 review of the field called it nascent and urged patience before drawing conclusions or real-world applications. Whenever you read "red makes people," ask what was manipulated.

The review lists a lack of experimenter blinding, failure to screen out color-deficient participants, nonstandard exposure times and failure to specify the color at the spectral level. An uncontrolled red may also differ from its comparison in lightness and chroma, so the cause is unknown.

Color has three properties, and the review notes that most theorizing considers hue alone, and mostly one hue, red. A study that compares a red with a blue usually changes lightness and chroma too. Unless those are matched, you cannot credit a result to hue alone.

A 2007 series, as a later meta-analysis summarizes it, reported that a small amount of red before or on a test lowered scores, with Cohen's d from about -0.51 to -1.14. Those are large for such a slight manipulation. Very large effects from tiny interventions should raise suspicion before they raise excitement.

Across 67 effect sizes from 38 samples: anagrams d = -0.06, knowledge tests d = -0.04, reasoning d = -0.34, with a 95% interval from -0.61 to -0.06. Once publication bias was corrected for, nothing reliable remained. Treat the red effect as contested and probably far smaller than first claimed.

The same meta-analysis found this in its cumulative analysis, and close replications failed to reproduce the originals. Shrinking effects over time are a standard warning sign of publication bias. When only one famous paper is cited for a color claim, look for what came next.

In one 2012 experiment, red sped people walking toward an interview about dating and slowed those walking toward an interview about intelligence. It is a single experiment. A theory that lets context decide is also easy to save after any result, so ask for tests fixed in advance.

In Chinese stock markets, red shows a rising price and green a falling one. One 2014 experiment found red lowered students' cognitive test scores while the reverse held for stockbrokers. It rests on a single study in a field with replication trouble, and it is a good reminder that meanings are learned.

One 2016 study reported red helping on simple proofreading, and blue helping on difficult proofreading and on creative tasks. Each new moderator explains an earlier mismatch and adds a claim to replicate. Keep such results as hypotheses rather than as a guide for painting a classroom.

Among 232 children in three small-scale societies and one Australian city, sex differences in pink preference were absent in the three societies (d from -0.31 to 0.23), and large in the city (d = 1.24). This cuts against claims that the preference is built in. Three societies is still a small base.

Reviews describe a smooth curve with a maximum near blue and a minimum near dark yellow, reported since 1897. Infants look longest at blue and least at green-yellow when colors are highly saturated, which hints at a sensory root. Whether any of this is "universal" remains disputed, since the pattern is reported mainly for industrialized cultures and one study found cultural variation.

In a study of 88 people choosing most and least preferred colors, walls favored lighter colors and T-shirts darker, less chromatic ones. The choices varied widely between people. So a "favorite color" survey result does not transfer directly to a product, a room or a garment.

In a study of 20 people rating saturated lights from -5 to +5, blue scored -1.87, green +1.09, yellow +2.1 and red +3.75, and every color warmed with intensity. Yet a hotter star is bluer. So warm means an association and an appearance, not a temperature, and the sample here is small.

Berlin and Kay's list has 11 possible basic color categories, and the review that cites it suggests words develop for what people need to describe in their environment. Having no separate word does not show that a person cannot see a difference. Keep vocabulary and perception apart.

Research using 330 chips from the World Color Survey treats naming as optimizing that trade-off, and even simulated networks that communicate reach similar systems. That is a functional account of why languages have the words they have. It does not say which colors any culture values.