Colour Constancy and Chromatic Adaptation | Writ In Light | Writ In Light

Colour Lesson VIII

Colour Constancy: the eye that rewrites what it sees

The eye is not a camera. It is a part of the brain.

Edwin Land

Lesson VII showed that two surfaces can masquerade as one under a single lamp. Metamerism is a vulnerability of the low-dimensional code the eye sends forward. This lesson asks the reverse question: how does a single surface hold its identity when the lamp keeps changing?

A white page under noon sun. The same page under golden hour light. The same page under a blue overcast sky. The spectral power reaching your retina shifts dramatically each time, yet you still call the page white. Something between the photon and the percept is rewriting the signal.

The page that stays white

Take a sheet of paper outside at midday. The light is roughly balanced across the visible spectrum, and the paper reflects most of it back. Now wait for golden hour. The spectrum arriving at your eye tilts heavily toward longer wavelengths. Measured with a spectrophotometer, the light bouncing off the paper is objectively warm . Yet you perceive the sheet as white, perhaps tinged, but still fundamentally the same surface.

This is colour constancy : the visual system’s quiet insistence that objects have a stable identity, even as illumination rewrites the signal at the retina.

Colour constancy is not passive reception. It is active inference: the brain peeling the illuminant away from the surface to recover what it believes the object 'really' is.

The mechanism is imperfect. Under very strong coloured light (a nightclub, a sodium streetlamp) constancy weakens, and surfaces drift toward the hue of the source. But across the ordinary range of daylight, overcast, tungsten, and shade, the system holds remarkably well.

Von Kries and the cone scaling trick

Behind constancy sits a process called chromatic adaptation . The most enduring model of that process belongs to Johannes von Kries, writing at the turn of the twentieth century.

The idea is simple. Each class of cone photoreceptor (long, medium, short wavelength) adjusts its own gain independently. Under warm light, which stimulates the L cones heavily, those cones dial their sensitivity down. Under cool light, the S cones do the same. The result is a rebalancing: the three cone signals return toward the ratio they would produce under a neutral illuminant.

In linear algebra, this amounts to a diagonal matrix applied to the cone responses. Three numbers, one per channel, each scaling the gain. It is the oldest formal adaptation transform, and it remains a building block inside every modern appearance model.

Von Kries adaptation is gain control at the gate: each cone class rescales itself to normalise the illuminant it is bathed in.

The elegance is also the limitation. A diagonal scaling cannot account for everything the visual system does. But as a first pass, it captures the core gesture: turn the volume down where the signal is loudest.

Land’s retinex: colour without correction

Edwin Land, inventor of the Polaroid camera, was unsatisfied with simple gain models. In 1977 he published his retinex theory (the name fuses “retina” and “cortex”), proposing that the visual system computes colour not from the absolute light at a point but from the ratios of light across boundaries.

His experiments used “Mondrian” displays: abstract patchworks of coloured paper illuminated by three independent projectors. Land could adjust the projectors so that a given patch sent the exact same triplet of cone signals as a completely different patch under different lighting. Yet observers continued to see the patches as their “true” colours, not the colour the raw physics would predict.

The retinex account says: the brain walks the scene, comparing reflected light at each edge, building a lightness map in each waveband independently. Absolute flux is discarded. Ratios survive.

Retinex says the brain does not trust absolute light. It reads edges, compares neighbours, and builds colour from contrast.

This helps explain why a grey patch can look white or near-black depending on its surround. The raw signal is the same; the relational signal is not.

Shadows that rewrite squares

Edward Adelson’s checker shadow illusion, published in 1995, is constancy made visible. A checkerboard sits under a cylinder that casts a diagonal shadow. Two squares, one inside the shadow and one outside, are printed at the exact same luminance in the image. Yet one looks clearly dark and the other clearly light .

The brain is doing its job. It infers the shadow, discounts the dimming it would cause, and recovers what it believes the underlying surface reflectance to be. In the real world, that inference is almost always correct: shadows change illumination, not surfaces. The illusion only fools you because the image is flat and the shadow is painted, not cast by actual geometry.

The illusion is not that you are deceived. It is that your brain is solving the right problem with the wrong scene.

Adelson’s image is a reminder that constancy is not a passive filter. It is a bet the visual system places on the structure of the world, and the bet usually pays off.

Beyond von Kries: appearance models

Von Kries cone scaling is a thermostat: effective, minimal, one dial per channel. Modern colour appearance models try to capture more of what the visual system actually does.

CIECAM02 (and its successor CAM16) adds surround luminance, background adaptation, and a more sophisticated chromatic adaptation transform called CAT02. It predicts not just whether two colours match, but how a colour looks under specific viewing conditions: dim versus bright surround, small versus large field, different white points. It is the workhorse for cross-media colour reproduction, connecting print proofs to screen previews.

The Hunt model, developed by R. W. G. Hunt across decades of research, is the most physiologically grounded of the three. It models several stages of adaptation, nonlinear cone responses, and rod intrusion at low light levels. The price is complexity: many parameters, careful calibration.

The progression tells a story. Each model tries to close the gap between a simple mathematical correction and the layered, context-sensitive inference the brain performs in milliseconds.

Von Kries is a thermostat. CIECAM02 is climate control. The brain is weather itself.

White balance: the camera tries to be an eye

A camera sensor has no cones, no gain control, no cortex. It records whatever spectrum the lens delivers. Left uncorrected, a photograph taken under tungsten light looks orange ; one taken in open shade looks blue . The colours are technically accurate to the physics, but they feel wrong because they lack the correction your visual system would have applied.

White balance is the photographer’s (or the firmware’s) attempt to close that gap. The algorithm typically assumes the average reflectance across the scene is a neutral grey , then shifts the colour channels to make that average land on neutral. It is von Kries adaptation implemented in silicon.

The assumption breaks in scenes that are not average. A field of golden wheat. A blue ocean filling the frame. A red curtain dominating the background. In each case auto white balance overcorrects, pushing the dominant colour toward grey and draining the scene of the very quality that made it striking.

When auto white balance fails, you glimpse a world without colour constancy: every illuminant shift repaints every surface.

Shooting in raw and setting white balance manually is, in a sense, choosing to be the cortex yourself: deciding which light to discount and which to keep.

Closing thought

Lesson VII: light can break the match between two surfaces. Lesson VIII: the brain fights back, rebuilding stable colour from unstable signals. Metamerism is the crack in the code. Constancy is the patch.

Both lessons belong together. One reveals the fragility of a three-channel system. The other reveals the intelligence layered on top of it.

The eye does not passively receive colour. It negotiates, infers, and insists. The light changes, and the brain holds the line.

References

    • Edwin Land, “The Retinex Theory of Color Vision,” Scientific American (1977) - Johannes von Kries, chromatic adaptation theory (1902, summarised in Fairchild) - Mark D. Fairchild, Color Appearance Models - Edward H. Adelson, “Checker Shadow Illusion” (1995) - CIE, CIECAM02 colour appearance model - David Briggs, huevaluechroma.com - Josef Albers, Interaction of Color