A gloss room looks less colourful than it measures

A gloss finish takes colour out of the whole room kept a ledger of every flux between a room’s faces and found that a glossy finish on two coloured walls removes about a tenth of the chroma from the room’s reflected light. It said what the ledger could not settle. A room-wide loss is a loss in the stimulus, and a viewer standing in the room adapts to its average light — which ought to discount a shift that affects everything equally. Whether a gloss finish makes a room look less colourful, or only measure less colourful, is a question for an appearance model.

The loss is where adaptation cannot reach it #

A viewer adapted to a room discounts the colour the whole room shares and keeps the colour that differs between its faces. A gloss finish takes as large a share of the second as of the first, so to an adapted viewer the room loses a larger share of what it can see than the light loses.

• At an eggshell finish, roughness 0.2, the room’s reflected light loses 8.3 per cent of its chroma and the six faces’ mean chroma falls 8.6 per cent to a viewer adapted to the lamp.
• To a viewer adapted to the room’s own average light the faces lose 13.9 per cent, and at roughness 0.15, 15.6.
• How far the faces sit from the room’s average loses 11.0 per cent to the adapted viewer and 11.2 to the unadapted one — the part of the loss adaptation cannot touch is the same size either way.
• Fully adapted, the room’s average light has no chroma at all, glossy or matt, to under a hundred-billionth of a unit.

A room’s colour is two things #

Standing in a room with two green walls, a viewer sees six faces lit by a lamp and by each other. Each face returns a light of its own colour, and those six lights have an average — the light a viewer is surrounded by.

The ledger measured every face’s light against the lamp’s white, as a colorimeter would. That reading mixes two things. One is the colour of the room’s average light: a room with two green walls fills itself with green-tinged light, which the room is the illuminant found for exactly this geometry, and every face, even the neutral ones, carries some of that tint. The other is how far each face’s light differs from that average: the painted walls more green than the room, the neutral walls less.

A visual system treats the two differently, and this is the whole reason adaptation exists. The colour of the average light is information about the illuminant, and constancy is the default: an observer discounts it, so that a white wall looks white in a green-lit room. The differences between faces are information about the surfaces, and those are what adaptation is there to preserve.

The shared colour disappears #

Reading every face through CIECAM16 with the adopted white set to the room’s own average reflected light separates the two parts cleanly.

With complete adaptation the room’s average light comes out with a chroma of five parts in a trillion, in the matt room and the glossy one alike. That is the exact null the argument rests on: whatever colour the room’s light has, a viewer fully adapted to it sees its average as grey. With the degree of adaptation the model itself computes for an ordinary lit room, 0.94, a trace of 3.8 units remains, against 46.7 when the same average is read against the lamp’s white.

So the part of the gloss finish’s loss that belongs to the room’s average light — the ledger’s headline — is not what the adapted viewer sees at all. The room’s light is less green in the glossy room than in the matt one, and an adapted viewer sees both as neutral. What the viewer sees is everything else.

What the unadapted viewer sees #

A viewer who has just walked in from daylight has not yet adapted to the room, and reads its faces against a white close to the lamp’s.

To that viewer the painted walls fall from a chroma of 93.2 to 82.2, the ceiling from 39.5 to 37.6, the floor from 24.1 to 22.9, and the back and front walls from 24.1 to 23.9. The mean over the six faces falls 8.6 per cent, almost exactly the share the ledger found in the light. The neutral walls are coloured at all only because the room’s green light reaches them, and the finish barely changes how much of it they receive.

What the adapted viewer sees #

Once the viewer has adapted to the room’s own light, the neutral walls are no longer tinted green. They are tinted against green.

In the matt room, to the adapted viewer, the painted walls have a chroma of 54.5 at a green hue of 144 degrees, and the back, front and floor 25.1 at a hue of 328 degrees — magenta, the opposite of the room’s light. The ceiling, which receives light close to the room’s average, is nearly neutral at 3.2. A wall painted neutral looks pinkish in a green room to an eye that has discounted the green, for the same reason a corner moves both terms: the viewing condition has changed, not the surface.

In the glossy room the painted walls fall to 45.9 and the neutral walls to 21.4. Every face except the ceiling loses a larger share of its chroma than it did to the unadapted viewer, and the back and front walls lose fifteen per cent where before they lost one. The mean over the six faces falls 13.9 per cent.

Why the share grows #

The two viewers disagree about the percentage and agree about the thing it is a percentage of.

The part adaptation removes is the average’s colour. Take it away and each face’s chroma becomes its distance from the room’s average. For the painted walls and the ceiling that is much smaller than their distance from the lamp’s white, because the room’s light already carried most of their colour; the neutral walls, which had borrowed the room’s tint, trade it for the opposite tint at about the same size. The mean over the six faces falls from 49.7 to 31.3, and the gloss finish shrinks the faces’ differences by about the same amount whichever white they are measured from — and the same amount out of a smaller whole is a larger share.

The quantity that shows this directly is the spread: how far, on average, the six faces’ appearances sit from the room’s average light, in the model’s uniform space. It measures only the uneven part. At roughness 0.2 the spread falls 11.2 per cent read against the lamp and 11.0 per cent read against the room. Across finishes from roughness 0.8 to 0.15 the two readings never differ by more than two tenths of a point, while the adapted viewer’s loss in the faces’ chroma runs from 5.8 to 15.6 per cent and the unadapted viewer’s from 3.4 to 9.4.

So the gloss finish takes about the same share of the room’s between-face colour as the ledger found it takes of the light — a little more — and adaptation, far from discounting that, strips away the shared colour that was making the loss look small.

The neutral walls, in the light and to the eye #

The ledger had its own face-by-face reading, and the neutral walls are where it and the adapted viewer part company most.

In the light, the back and front walls receive light 7 to 8 per cent less colourful at roughness 0.2 and send out light 0.2 per cent less colourful — they are neutral, so the colour they send out is what the room hands them, and the room hands them nearly as much green as before. To a colorimeter they are the faces the finish changes least. To a viewer adapted to the room they are among the faces it changes most, losing 14.8 per cent of their apparent chroma.

The two readings do not contradict each other. The ledger asks how green the light leaving the back wall is, and the answer barely moves. The adapted viewer asks how far that light is from the room’s average, and the room’s average has moved: in the glossy room it is less green, while the back wall’s light is nearly as green as it was. A wall that keeps its colour while its surroundings lose theirs looks less different from them, and to an eye that reads colour against its surroundings, less different is less colourful. The neutral walls’ borrowed magenta is the purest case, because none of it belongs to the wall.

That is the mechanism of the whole result in one pair of faces. The ledger’s measure moves with each face’s own light; the viewer’s moves with each face’s light against the room’s, and a finish that changes the room’s light changes every face’s standing in it.

How far the viewer has adapted #

The viewers above are two ends of a range. A real one adapts partly, and to a white somewhere between the lamp’s and the room’s.

With the adopted white a quarter of the way to the room’s the faces lose 9.1 per cent; halfway, 9.6; three quarters, 14.0; all the way, 13.9. The spread loses between 11.0 and 12.2 per cent at every step. The rise is not smooth, because at the halfway white the neutral walls’ light happens to match the adopted white almost exactly, their chroma falls to under two units, and a face with no chroma cannot lose a share of it.

A discount nobody measured put the degree of adaptation in an ordinary room at 0.94 by the model’s own formula, and the question of which white a viewer adapts to in a room of differently coloured surfaces has no such formula at all. The result does not need one. Whatever white is adopted, the uneven part loses the same share, and the only thing adaptation changes is how much shared colour is left in the figure it is a share of.

Three paints #

The ledger found the room’s loss nearly the same for blue, green and orange walls, and the walls’ own losses very different. The adapted viewer’s losses follow neither pattern exactly.

Blue walls lose 9.9 per cent of the light’s chroma, 7.3 per cent of the faces’ chroma to the unadapted viewer and 14.1 per cent to the adapted one. Green walls lose 8.3, 8.6 and 13.9; orange walls 9.0, 13.0 and 20.2. The spread loses 11.7, 11.0 and 16.9 per cent to the adapted viewer. For all three colours the adapted viewer sees the larger loss, and for orange walls — whose room light is least coloured to begin with, so that less of their colour is shared — the between-face loss is largest of all.

The paint that the ledger said made least difference to the room’s light makes a large difference to what an adapted viewer sees: an orange room in eggshell looks a fifth less colourful than the same room in flat paint.

What the light said and what the eye says #

The ledger’s own reading of the room is still correct, and it is worth setting beside the viewer’s to see what each is a reading of.

The light loses 3.0 per cent of its chroma at roughness 0.8 and 9.1 per cent at 0.15, while its quantity rises by up to a fifth. A colorimeter reading the room’s light reports the first series, and a viewer adapted to the room sees the faces lose 5.8 and 15.6 per cent. Matching is not appearance, and here the difference has a sign: the measurement reports between a half and three fifths of the loss a person adapted to the room would see.

That is not a flaw in the ledger. It measured the light, and the light’s chroma includes the colour every face shares — the colour an observer throws away first. A reader deciding whether a finish will change how a room looks needs the part that is kept.

What a finish chart cannot show #

A paint finish is chosen from a sample card, and the essay on the ledger noted that a card is a single bounce with nothing to interreflect. The appearance calculation adds a second reason the card misleads. On the card, the colour a viewer judges is the paint against the white of the card’s border, which is close to the lamp’s white; in the room, the colour a viewer judges is the paint against the room’s own light, which the paint has tinted.

So a flat and an eggshell finish of the same paint, compared on cards under the same lamp, differ in chroma by whatever their single-bounce reflections differ by. Compared in finished rooms by a person who has adapted to each, they differ by more than the light in those rooms does, because the shared tint that made both rooms look alike has been discounted from both. A designer who wants a coloured room to keep its colour in a glossy finish has to add more colour than the card suggests — and more than a colorimeter in the room would suggest too.

How the appearances were computed #

The rooms are the unit cube of the ledger: a lamp in the ceiling emitting D65, two opposite walls painted with a Gaussian reflectance band of width 25 nm and peak albedo 0.85, the other faces a neutral 0.5, every face carrying a microfacet lobe of stated roughness, solved directionally in eighty-one bands. Each face’s outgoing reflected flux, with the lamp’s own emission removed, is its light.

Each face’s light is converted to tristimulus values and scaled so that the mean over the six faces has a luminance factor of 20, a mid grey. It is read through CIECAM16 at an adapting luminance of 100 candelas a square metre, a background of 20 and an average surround, with an adopted white that is the lamp’s own, the room’s total reflected light, or a mixture of the two at unit luminance; the degree of adaptation is the model’s own unless it is set to one. The faces’ chroma is averaged over the six faces with equal weight, since each fills the same part of a viewer’s surroundings; the spread is the root-mean-square distance of the six faces from the room’s average light in CAM16-UCS’s opponent coordinates.

What the calculation assumes #

The viewer adapts to the room’s average reflected light, weighted by area. A viewer who looks mostly at one wall, or who has the lamp itself in view, adapts to something else, and the mixture sweep brackets that without identifying it.

Adaptation is global. A patch is not a scene, and a real visual system also adapts locally — a large green wall in view shifts the adaptation for what is near it — which would add structure to the face-by-face reading that one adopted white cannot represent.

And every face is read as a uniform patch at a mid-grey level. The faces of a real room have gradients across them, the corners are brighter, and the painted walls are seen at many angles; the floor is a different colour from the door, which a single outgoing flux per face averages away.

The habit: an adapting observer keeps the differences #

A shift that affects everything alike is discounted by any system that normalises to its surroundings, and the obvious conclusion is that such a shift matters less than it measures. The conclusion is right for the shift and wrong for anything measured alongside it, because normalising removes the shared part from every quantity at once.

The move is to split a change into what the whole field shares and what varies across it before asking what an adapting observer makes of it. The shared part goes; the varying part stays; and whether the remaining change looks larger or smaller depends on how much of the original quantity was shared.

The failure mode is to reason from the discount to a smaller effect. Discounting the shared colour of a room does not shrink the loss of its colour — it shrinks the colour, and leaves the loss.

Where this comes from #

Chromatic adaptation to the average of a scene is the grey-world assumption as visual systems implement it, and the algorithms that guess the light describe the same assumption in cameras. That interreflections between coloured surfaces tint a room’s light and that observers partly discount the tint is part of the literature on colour constancy in real rooms.

That a gloss finish’s loss of colour is larger to a viewer adapted to the room than in the room’s light, that the between-face part of the loss is the same size whether or not the viewer has adapted, and that neutral walls in a green room look magenta to an adapted eye with the finish reducing that too, are computed here with CIECAM16 on the directional room solution.

Still open: whether people see it #

The model says what an observer adapted to a room’s average light would report, and it was fitted to judgements of patches in controlled surrounds rather than of rooms. Two physical rooms, or two rendered ones shown immersively, identical except for the finish on their painted walls, with observers asked to match the colour of each wall after adapting to each room in turn, would say whether the finish costs a room a tenth of its colour or nearly a sixth — and whether the neutral walls’ borrowed magenta is something people notice at all.