How Colour Is Measured: From the Eye to Numerical Values

Turning a sensation into a number

Colour begins as a private event. Light enters the eye, three classes of cone respond, the visual cortex constructs an experience, and the observer says the sample is a slightly greenish blue. None of that transfers reliably down a telephone line or across a supply chain. Colorimetry exists to convert the private event into public numbers, so that a shade agreed in one country can be reproduced in another without either party seeing the other’s sample.

The achievement is more surprising than it sounds. The system does not measure colour in any physical sense, because colour is not a physical quantity. It measures light, then applies a model of human vision to predict what an average observer would experience. The numbers are therefore statements about people as much as about materials, and understanding where the model comes from explains most of its behaviour in practice.

The standard observer

In the late 1920s, W. David Wright and John Guild independently ran colour matching experiments. Participants viewed a split field: on one side a test colour of a single wavelength, on the other a mixture of three fixed primaries whose intensities the participant adjusted until the two halves matched. Repeated across the spectrum and across observers, the results gave the amount of each primary needed to match every wavelength.

The Commission Internationale de l’Eclairage combined and transformed these data into the colour matching functions known as x̄, ȳ and z̄. Published in 1931, they define the CIE standard observer: a statistical stand in for normal human colour vision. The 1931 functions describe a 2 degree field of view, matching the small central patch used in the original work. A 10 degree set followed in 1964 for larger samples, and it is the better choice for most industrial assessment, where a swatch fills far more of the visual field than two degrees.

From three curves to three numbers

With the observer defined, measurement becomes arithmetic. Take the spectral power distribution of the illuminant, multiply it wavelength by wavelength by the spectral reflectance of the sample, multiply again by each of the three colour matching functions, and integrate across the visible band. The result is three numbers, the tristimulus values X, Y and Z.

Y is constructed to correspond to luminance, since ȳ deliberately matches the eye’s luminous efficiency curve. X and Z carry the chromatic information. Together they specify, for a given illuminant and observer, everything the visual system can extract from that surface. Two samples with the same tristimulus values under the same conditions are, by definition, an exact visual match. Whether they remain one under a different illuminant is the metamerism question, and the tristimulus values alone cannot answer it.

The trouble with XYZ, and the rise of CIELAB

XYZ is complete but hostile to intuition. The values do not correspond to anything an observer would say about a colour, and worse, the space is perceptually non uniform. Equal numerical distances in XYZ correspond to wildly unequal perceived differences. David MacAdam demonstrated this in the 1940s by mapping the regions within which observers could not distinguish a colour from a reference. Plotted on the CIE chromaticity diagram, these regions form ellipses of dramatically different size and orientation, large and elongated in the greens, small and tight in the blues. A tolerance expressed as a distance in that space means something different in every region.

The response, formalised in 1976, was CIELAB, sometimes written L*a*b*. It applies a cube root transformation to the tristimulus values, normalised against a reference white, to produce three axes intended to be perceptually more even.

  • L*. Lightness, running from 0 for absolute black to 100 for the reference white.
  • a*. The red to green axis. Positive values are redder, negative values greener.
  • b*. The yellow to blue axis. Positive is yellower, negative is bluer.

The opponent structure is not arbitrary. It reflects how the visual system encodes signals after the cones, combining cone outputs into a lightness channel and two chromatic difference channels. It also explains why there is no such thing as a reddish green: the axis has one dimension, so a colour sits somewhere along it and cannot be at both ends.

The cylindrical view

Many practitioners prefer the polar form, CIELCh, which describes the same space in terms an operator recognises. C* is chroma, the distance from the neutral axis, corresponding to how saturated or vivid the colour appears. h is hue angle, measured in degrees around the a*b* plane, where roughly 0 is red, 90 yellow, 180 green and 270 blue. Saying that a batch is two units low in chroma and one degree off in hue is immediately actionable in a way that a shift in a* and b* is not.

Delta E: putting a size on a difference

Communicating a colour is only half the task. The other half is agreeing how far a sample may stray before it is rejected. That is the role of Delta E, written ΔE, a single figure expressing the magnitude of the difference between two colours.

The original 1976 formula, ΔE*ab, is the straight line distance between two points in CIELAB: the square root of the summed squares of the differences in L*, a* and b*. Its appeal is simplicity. Its weakness is that CIELAB turned out to be less uniform than hoped, particularly in saturated regions, so a ΔE of 2 in one part of the space does not look like a ΔE of 2 elsewhere.

The successors

  • CMC(l:c). Developed by the Colour Measurement Committee of the Society of Dyers and Colourists, it builds an ellipsoidal acceptability volume around the standard, sized according to where in the space the standard sits. The l and c parameters weight lightness against chroma; CMC(2:1) is the textile default, tolerating lightness variation more readily than hue error, which mirrors what people actually accept in a garment.
  • ΔE94. A CIE formula applying weighting functions to chroma and hue, with separate parameter sets for graphic arts and for textiles.
  • ΔE00, or CIEDE2000. The current recommendation. It adds a hue dependent rotation term to handle the troublesome blue region, plus compensations for lightness, chroma and hue, and a correction for the neutral axis. The formula is unpleasant to compute by hand and correlates best with visual judgement, which is the only test that matters.

As a rough orientation, a ΔE00 below 1 is generally imperceptible to a trained observer viewing samples side by side. Between 1 and 2 the difference is perceptible on close comparison but often acceptable. Above 3 it is usually obvious. These figures are guidance, not law: tolerances belong to the product. A pack of premium cosmetics and a workshop floor coating do not deserve the same limit, and neither does a shade of navy compared with a pale grey.

Instruments and the conditions that decide the answer

Two instrument families dominate. A spectrophotometer disperses the reflected light and records reflectance in narrow bands, typically every 10 nanometres, producing the full curve from which any illuminant and observer combination can be computed. A colorimeter uses three filtered detectors approximating the colour matching functions and returns tristimulus values directly. Colorimeters are cheaper and quicker; they cannot detect metamerism, because they never see the spectrum.

Measurement geometry is not a detail. The two common arrangements are 45/0, which illuminates at 45 degrees and views perpendicularly, excluding the specular reflection and so approximating how a person looks at a sample, and d/8, which uses an integrating sphere and can include or exclude the specular component. On a glossy surface the two geometries disagree substantially, and neither is wrong. Numbers from different geometries are simply not comparable, and any specification that omits geometry has left a hole a supplier can drive through.

  • Aperture size. A small aperture on a textured or heathered material samples too little to be representative. Use the largest aperture the sample allows, and average several readings taken at different orientations.
  • Backing and opacity. Thin or translucent samples transmit light to whatever sits behind them. Fold fabric to opacity, or use a defined backing consistently at both ends of the supply chain.
  • UV content. Instruments offer UV included, UV excluded and UV calibrated modes. On material containing optical brighteners the three give materially different answers, and the mode must be stated.
  • Verification. White tile calibration checks the instrument against itself. A set of coloured ceramic tiles measured periodically and tracked over time is what actually reveals drift, and it is what allows two laboratories to establish inter instrument agreement before they start arguing about a batch.

None of this replaces the eye. It disciplines it. The number tells the dyehouse which direction to move and by how much; the visual assessment under standardised lighting confirms that the number has been interpreted sensibly. A colour programme that trusts only the instrument will ship matches that look wrong, and one that trusts only the eye will never settle an argument.

Read next: The Periodic Table Color-Coded: How Element Groups Get Their Hues