Observer Metamerism

Two people stand at the same booth, look at the same pair of samples, and disagree about whether they match. Both are looking carefully. Both are telling the truth. This is the one colour disagreement no measurement will settle.

The standard observer is an average

Every colour number rests on a description of human vision: how sensitive the eye is at each wavelength. It came from experiments in the 1920s and 1930s in which a few people matched colours in a laboratory, and their results were averaged into the standard observer.

It has served the industry well. It is also an average of a handful of people who are long dead, and nobody’s eyes are exactly it.

Why real eyes differ

None of this is a defect. It is the normal range of normal colour vision, and separate from colour vision deficiency.

When it becomes a problem

Almost never, when two surfaces have similar spectra. If a sample and its reproduction return light in much the same shape, everybody agrees, because there is nothing for individual differences to get hold of.

It becomes a problem exactly where metamerism already lives: two surfaces with very different spectra, matched to the same numbers.

A metameric pair is balanced on a knife edge. It matches for the standard observer, which is to say for an average of people who are not in the room. Move the sensitivity curves slightly — a different pair of eyes, a different age — and the balance tips. One person sees a match; the next sees a mismatch in a definite direction, and can usually tell you which way.

Where it turns up most today

Screen against print. A display makes colour from three narrow bands of light; ink makes it from broad, smooth reflectance. Those two spectra could hardly be less alike, so a screen-to-print match is a metameric match almost by construction — and therefore the most exposed to individual variation. Two colleagues comparing a proof to a monitor are the classic case, and neither of them is being difficult.

Narrow-band lighting. LED sources are getting narrower and spikier, which sharpens the same effect. A booth that sits exactly on the right white point can still make two people disagree, because the shape of the spectrum matters as much as where its average lands.

What to do about it

Prefer spectral matches over numerical ones. If the two surfaces have similar curves, individual differences have nothing to work with. This is the reason a colour specification carries a spectrum in the first place — not to be thorough, but to make the match robust against the person looking at it.

Do not settle a disagreement by asking a third person. If two careful people disagree about a metameric pair, the answer is not another opinion. It is to change the pair, or to accept that it will look different to different customers.

Use the right observer for the size. The 2° observer describes a small patch; the 10° one describes a large field. Choosing the wrong one is not observer metamerism, but it produces the same kind of argument and it is entirely avoidable — see Light, Spectrum, Observer.

Judge critical colour under critical conditions. A proper viewing condition will not abolish individual variation, but it removes every other variable so that what remains is genuinely this and not something simpler — see Measuring and Viewing.

Can it be measured?

Yes, in the sense that matters. Rather than one average observer, colour science has sets of observers that represent the range real people fall into, and a pair of colours can be evaluated against all of them instead of one. The result is an index of how much a given match is likely to be disputed — not a prediction about any individual, but a fair statement of the risk.

ChromaSpot holds the data for that work and does not publish it yet. The tables belong to somebody else’s published research and have not been checked line by line against the paper they came from, and they cover a narrower range of wavelengths than this library does. Both are solvable; neither is solved. Publishing someone else’s tables under our name without verifying them would be exactly the kind of shortcut this library exists not to take.