The human visual system's tendency to perceive an object's colour as relatively stable even when the lighting around it changes.
The brain partially compensates for changes in illuminant colour temperature so that a white object still looks roughly white under both warm and cool light — this compensation is imperfect and varies between observers, which is part of why people can genuinely disagree about what colour something (including an eye) looks.
Colour constancy is a perceptual/brain effect, not a property of the object itself — it's a different phenomenon from metamerism, which is about the physical light reaching the eye rather than how the brain interprets it.
Two observers looking at the same eyes under the same light can genuinely disagree on whether they read as green or hazel, partly because of individual differences in colour constancy.
Explains why eye-colour classification is inherently somewhat subjective, even before cameras and photography are involved.
Constancy is studied by asking observers to match or name a surface under different lights and comparing their answers with what the changed light alone would predict. In calculation it is represented by a chromatic adaptation transform, such as Bradford or the CAT16 transform inside CIECAM16, which converts tristimulus values under one white into the values that would look the same under another. The transform is a model of an average observer who has had time to adapt, and real viewers fall short of it in ways that depend on the scene.
| Light | Adapted appearance | ΔE00 from its daylight appearance |
|---|---|---|
| Daylight, CIE D65 #e2725b | #e2725b | 0.0 |
| Daylight, CIE D50 (graphic arts viewing) #e4755a | #e4755a | 1.4 |
| Incandescent tungsten, CIE A #e97f55 | #e97f55 | 6.5 |
| Narrow-band tri-phosphor fluorescent, CIE FL11 #e9775d | #e9775d | 1.9 |
| Red-green-blue mixed LED, CIE LED-RGB1 #fa615a | #fa615a | 6.0 |
Fully adapted to tungsten light, the terracotta surface still looks 6.5 ΔE00 different from how it looks in daylight, and under that lamp it shifts most. Adaptation removes most of the change in the light and never all of it. The remainder is real, and it is what people mean when they say a colour 'changes' in the evening.
How this was calculated: A modelled reflectance for Terracotta #e2725b under each CIE illuminant, with Bradford chromatic adaptation standing in for the visual system. Human constancy is less complete than this transform in dim or strongly coloured light.
A photograph of a dress, a wall or a product is viewed with no clue to the light it was taken in.
Viewers discount whichever light they assume, and two people who assume differently see two different colours in one image. Including a neutral reference, or the surroundings, in the frame removes the ambiguity.
A wall colour is judged after sitting in the room for some time.
Adaptation absorbs the cast of the room's light, so a strongly tinted lamp seems neutral and the paint is blamed or praised for what the lamp is doing. Step outside and back in, or compare against a white card, before deciding.