Colour Constancy

match what the colour is, not what it looks like

round: –
score: –
best: –

Loading…

the patch

yours

white card grey block prop

the scene — one coloured light over everything

the big square is the patch. the white card is really white — whatever colour it looks is the light. take that away from the patch, then build what is left on the right.

yours

true

your rebuild — the sliders paint this chip, on honest grey

this side has no coloured light on it at all. paint the patch as it would look under plain daylight.

hue 180°

sat 40%

light 55%

One coloured light tints the whole scene — the ground, the white card, the grey block, the prop and the patch. The patch is the big square. Your job: work out what colour the patch really is — what it would look like under plain daylight — and build that on the right with the sliders, then lock it in. Four colours per round. The ± buttons nudge a slider by exactly 1, and so do the arrow keys; hold a ± button down and it walks, which is how you cover ground on the hue slider without a hundred taps.

The trick: the card labelled “white card” is genuinely white. So whatever colour it appears to be is the colour of the light. Take that away from everything else and what is left is the real colour. Painters call this discounting the light; your eye already does it for you in real life, which is exactly why it is hard to switch off when you are painting.

Scoring weighs the lesson far heavier than your slider hand. The bit of your miss that lies along the light's own direction — the light you failed to take off — is charged against the cast itself: leave a fifth of the light in and you drop about a fifth of the marks, and on a strong cast copying what you see scores nothing at all. On the gentlest cast a copy can still take up to half marks on that one colour — a light that barely tinted the patch cannot be worth as much to take off as one that shouted, so the drill will not pretend it was; copy every colour in a round and the round still comes out around 10. Ordinary slider slop across that direction costs less than a third as much. So understanding the light and being a few degrees out beats being precise about the wrong colour. Dead-on is still exactly 100. On reveal the light switches off so you can watch the cast fall away, and the readout says how much of the light you left in, which way you leaned (hue / sat / light) and the raw ΔE. The round ends with your average lean.

One honest warning, because it is a hand problem and not an eye problem: slop that happens to point along the light gets charged like light you left in — from the outside the two are the same miss. On a phone the hue track carries all 360 degrees across a couple of hundred pixels, so a thumb that lands four pixels off has landed six or seven degrees off, and that can cost real marks even when the patch already looks right. Land the last few degrees with the ± buttons or the arrow keys instead: they move by exactly one, on every device, and holding a ± walks it.

Studio tip: the sliders start where the patch looks, not on some unrelated colour. Your job is to move away from what you see — which is the whole lesson.

Seeing past a coloured light

One coloured light tints a small scene: the ground, a labelled white card, a grey block, an off-hue prop and the patch itself. The scene is built as reflectance times illuminant per channel in linear RGB, so the light only removes energy and nothing clips. On a neutral grey field beside it you rebuild the patch's real surface colour with hue, saturation and lightness sliders — ±1 buttons for the last few degrees, held to walk — then lock it in. Four colours a round, the light strengthening and the patch growing more muted as it runs.

The card labelled white is genuinely white, so whatever colour it appears to be is the colour of the light. Subtract that and the surface remains. Your eye does this discounting all day unasked, and that is the reflex a painter has to switch off. Deals where the light barely touches the patch are re-dealt: an item you can win by copying what you see teaches nothing.

How much of the light you left in

Each lock is measured as CIE76 ΔE against the patch's true reflectance and mapped through 100 × clamp(1 − ΔE/30, 0, 1). The true reflectance is a number the scene was built from, not something a model has to guess at, so that formula is the whole judge. The miss is then split by direction: the part along the light's own axis — light you did not take off — is charged in full, while slop across that axis costs under a third as much. Knowing the light and being a few degrees out beats precision about the wrong colour; on a strong cast, copying what you see scores nothing. That ΔE 30 zero point is identical on every device; what a coarse pointer costs you is taps on the ±1 buttons, not marks. On reveal the light drops back to neutral, so the cast visibly leaves the scene, and your hue, saturation and lightness misses print with their signs; the round is the mean of the four, and closes with the way your eye leans.

🎭 Colour Constancy

Seeing past a coloured light

How much of the light you left in

Colour Constancy