Short answer
Most camera sensors are colour-blind silicon under a mosaic of red, green and blue filters, usually in the Bayer pattern of two greens for every red and blue. Software fills in the missing two colours at each pixel and converts the result to a standard colour space. But the filters' spectral sensitivities are not the eye's, so a camera can see two colours as matching when you see them as different, and the reverse. A camera is a good estimator of colour, not a colorimeter.
A photosite counts photons; it has no idea of their wavelength. To record colour, almost every consumer camera covers its sensor with a colour filter array, one tiny red, green or blue filter per photosite. The most common layout was patented by Bryce Bayer at Eastman Kodak, filed in 1975 and granted in 1976: green filters at every other position, with red and blue sharing the rest. Green gets double the sites because it carries most of the brightness detail the eye relies on. Each pixel therefore measures one colour, and the camera estimates the other two from neighbours, a step called demosaicking.
Demosaicking works by assuming colour changes more slowly than brightness across an image. Where that assumption fails — fine repeating patterns, sharp edges between saturated colours, hair and fabric texture — the interpolation can produce false colour: rainbow fringes on fine stripes or coloured speckles at high-contrast edges. Better algorithms reduce it, and many cameras once used an optical low-pass filter to blur detail slightly before it reached the mosaic. It is a sampling artefact of recording one colour per pixel, not a property of the scene.
Human colour vision is defined, for measurement purposes, by the CIE colour-matching functions derived from the eye's three cone types. A camera would record colour exactly as a standard observer sees it only if its three channel sensitivities were a linear combination of those functions — the Luther condition. Real filters are chosen for noise, sensitivity and manufacturability as well, and they fall short. A study that measured 28 cameras, from professional SLRs to a phone, found most deviate from the Luther condition, some substantially. The consequence is camera metamerism: two surfaces that match for the eye can be recorded differently, and two that the camera records identically can look different to you.
This is why a colour-correction matrix, however carefully built, is a best fit across typical colours rather than an exact conversion.
Raw sensor values are in the camera's own colour space, defined by its filters. The processing chain scales the channels so that neutral objects come out neutral (white balance), demosaics, and then applies a 3×3 colour-correction matrix to map camera RGB into a standard space such as linear sRGB, before tone curves and gamma encoding. Every one of those steps is an estimate tuned for common scenes. Colours far from the training set — very saturated LED light, fluorescent inks, some dyes — are where a camera's colour is least trustworthy.
Why: Demosaicking false colour where fine detail approaches the filter-mosaic spacing.
Fix: Change distance or zoom slightly; in raw editing, use a false-colour or moiré reduction tool.
Why: Camera metamerism: the sensor's sensitivities differ from the eye's.
Fix: Check under the actual light with the eye; do not rely on a photo for a match.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
FactStrong evidence
The Bayer colour filter array (US Patent 3,971,065, filed 1975 and granted 20 July 1976 to Eastman Kodak) places green, luminance-sensing elements at every other position, with red and blue elements filling the remainder.
Source: US Patent 3,971,065: Color imaging array (Bryce E. Bayer, Eastman Kodak Co.)
FactStrong evidence
A camera satisfies the Luther condition if its spectral sensitivity functions are a linear transformation of the CIE 1931 2° colour-matching functions; measurements of 28 cameras found that most deviate from it.
Caveat: The measured cameras date from before 2013; newer sensors may differ in degree but not in kind.
StandardStrong evidence
The CIE standard observer and cone-fundamental functions define colour matching for human vision, which is the reference a colorimetric camera would have to reproduce.
Source: CIE 015:2018 Colorimetry, 4th edition; CIE 170-1:2006 Fundamental Chromaticity Diagram with Physiological Axes – Part 1
FactStrong evidence
A published smartphone pipeline converts sensor RGB to linear sRGB with a 3×3 colour-correction matrix after white balancing and demosaicking.
Reviewed 29 September 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.