David H. Foster

For a stable visual world, the colours of objects should appear the same under different lights. This property of colour constancy has been assumed to be fundamental to vision, and many experimental attempts have been made to quantify it. I contend here, however, that the usual methods of measurement are either too coarse or concentrate not on colour constancy itself, but on other, complementary aspects of scene perception. Whether colour constancy exists other than in nominal terms remains unclear.

A quarter of a century ago, the first systematic behavioral experiments were performed to clarify the nature of color constancy—the effect whereby the perceived color of a surface remains constant despite changes in the spectrum of the illumination. At about the same time, new models of color constancy appeared, along with physiological data on cortical mechanisms and photographic colorimetric measurements of natural scenes. Since then, as this review shows, there have been many advances. The theoretical requirements for constancy have been better delineated and the range of experimental techniques has been greatly expanded; novel invariant properties of images and a variety of neural mechanisms have been identified; and increasing recognition has been given to the relevance of natural surfaces and scenes as laboratory stimuli. Even so, there remain many theoretical and experimental challenges, not least to develop an account of color constancy that goes beyond deterministic and relatively simple laboratory stimuli and instead deals with the intrinsically variable nature of surfaces and illuminations present in the natural world.

Naive observers viewed a sequence of colored Mondrian patterns, simulated on a color monitor. Each pattern was presented twice in succession, first under one daylight illuminant with a correlated color temperature of either 16,000 or 4,000 K and then under the other, to test for color constancy. The observers compared the central square of the pattern across illuminants, either rating it for sameness of material appearance or sameness of hue and saturation or judging an objective property—that is, whether its change of color originated from a change in material or only from a change in illumination. Average color constancy indices were high for material appearance ratings and binary judgments of origin and low for hue–saturation ratings. Individuals’ performance varied, but judgments of material and of hue and saturation remained demarcated. Observers seem able to separate phenomenal percepts from their ontological projections of mental appearance onto physical phenomena; thus, even when a chromatic change alters perceived hue and saturation, observers can reliably infer the cause, the constancy of the underlying surface spectral reflectance.

Small changes in daylight in the environment can produce large changes in reflected light, even over short intervals of time. Do these changes limit the visual recognition of surfaces by their colour? To address this question, information-theoretic methods were used to estimate computationally the maximum number of surfaces in a sample that can be identified as the same after an interval. Scene data were taken from successive hyperspectral radiance images. With no illumination change, the average number of surfaces distinguishable by colour was of the order of 10,000. But with an illumination change, the average number still identifiable declined rapidly with change duration. In one condition, the number after two minutes was around 600, after 10 min around 200, and after an hour around 70. These limits on identification are much lower than with spectral changes in daylight. No recoding of the colour signal is likely to recover surface identity lost in this uncertain environment.

Shepard's analysis of how shape, motion, and color are perceptually represented can be generalized. Apparent motion and shape may be associated with a group of spatial transformations, accounting for rigid and plastic motion, and perceived object color may be associated with a group of illuminant transformations, accounting for the discriminability of surface-reflectance changes and illuminant changes beyond daylight. The phenomenological and mathematical parallels between these perceptual domains may indicate common organizational rules, rather than specific ecological adaptations. [Barlow; Hecht; Kubovy & Epstein; Schwartz; Shepard; Todorovic].

An important factor in judging whether two retinal images arise from the same object viewed from different positions may be the presence of certain properties or cues that are 'qualitative invariants' with respect to the natural transformations, particularly affine transformations, associated with changes in viewpoint. To test whether observers use certain affine qualitative cues such as concavity, convexity, collinearity, and parallelism of the image elements, a 'same-different' discrimination experiment was carried out with planar patterns that were defined by four points either connected by straight line segments (line patterns) or marked by dots (dot patterns). The first three points of each pattern were generated randomly; the fourth point fell on their diagonal bisector. According to the position of that point, the patterns were concave, triangular (three points being collinear), convex, or parallel sided. In a 'same' trial, an affine transformation was applied to one of two identical patterns; in a 'different' trial, the affine transformation was applied after the point lying on the diagonal bisector was perturbed a short, fixed distance along the bisector, inwards for one pattern and outwards for the other. Observers' ability to discriminate 'same' from 'different' pairs of patterns depended strongly on the position of the fourth, displaced, point: performance varied rapidly when the position of the displaced point was such that the patterns were nearly triangular or nearly parallel sided, consistent with observers using the hypothesised qualitative cues. The experimental data were fitted with a simple probabilistic model of discrimination performance that used a combination of these qualitative cues and a single quantitative cue.

The aim of this study was to test red - green dichromats' ability to discriminate between illuminant and surface-reflectance changes in natural scenes. Stimuli were simulations of natural scenes presented on a colour monitor with 10-bit resolution per gun. The natural scenes were obtained with a fast hyperspectral imaging system. Six different scenes (including rocks, foliage, and buildings) were tested. In each trial, two images were presented in sequence, each for 1 s, with no interval. The images differed in the phase of daylight on the scene: first with correlated colour temperature 25 000 K, then 6 700 K. The spectral reflectance of a region in the second image was changed randomly, consistent with a local change in daylight. The observer's task was to decide whether a particular surface in the successive images was the same (pure illuminant change) or different (illuminant change with a surface-reflectance change). The performance of four deuteranopes and five protanopes varied considerably across the scenes tested, with mean colour constancy indices (± 1 SEM) of 0.48 (± 0.13) and 0.19 (± 0.10), respectively (1.0 representing ideal performance). Deuteranopes seemed less disadvantaged than protanopes, performing close to normal with some scenes.

Colour constancy refers to the ability to extract information about surface colours independently of illumination conditions. A ripe strawberry, for example, appears the same red when viewed under a blue sky or a reddish sunset. Since Land's pioneering work, discussion has centred on the issue whether colour constancy is achieved primarily in the retina or visual cortex. Recently, the debate has shifted to a consideration of the constraints imposed by various psychophysical tasks and instructions. Humans can judge illuminant colour, reflected-light colour, surface colour, and the relationship between surface colours within a scene. Such observations suggest that colour constancy may not be a unitary phenomenon, and that its different aspects may be mediated at different levels in the visual system. Several questions therefore arise: are the tasks of surface estimation and illuminant estimation complementary, insofar as good performance in one implies poor performance in the other? How does retinal adaptation alter performance? Is colour vision tuned to the statistical composition of natural surfaces and illuminants? What role does colour-opponent processing in retina and cortex play in colour constancy? These and related questions are explored here in order to better understand what we mean when we ask: "Why do strawberries look red?"