Perception is remarkable in keeping things appearing much the same in varying conditions. Colours are hardly affected by changes of the ambient light, and sizes remain almost constant for different distances (though this breaks down for large distances, or looking downwards from a considerable height). Evidently there are active scaling processes that make the world appear stable under the ever-changing conditions of perceiving things. When the scaling is not appropriate many illusions are created.
1. Colour constancy
2. Chromatic adaptation
3. Models of colour constancy
4. Simultaneous colour contrast
5. Colour assimilation
6. Measurements of colour perception
1. Colour constancy
Colour is indeed a construct of the mind, but many modern scientists now believe that it is constructed for a specific purpose: to measure an immutable, physical property, which does exist 'out there'. We possess colour constancy, a fundamental mechanism of vision, which ensures that object colours remain nearly constant even under changing illumination. Thus, a yellow banana stays yellow whether we see it in sunlight or under the glow of a tungsten lamp.Colour constancy is a remarkable feat, because the light spectrum reflected by the banana changes markedly as the incident light spectrum changes. What remains invariant (until the fruit begins to spoil) is the spectral reflectance of the banana's surface, the physical characteristic that describes the proportion of incident light that it reflects at each wavelength, which is governed largely by the pigment embedded in its skin. Colour constancy enables us to compensate for changes in the spectrum of the illumination, and extract a measure of the banana's surface spectral reflectance from the changing light it reflects. How? We now know that there are several mechanisms that contribute to colour constancy.
2. Chromatic adaptation
At the lowest level of the visual system, chromatic adaptation plays a key role. Each of the three types of cone photoreceptor is responsive to a broad segment of the spectrum (the long- (L), middle- (M), and short- (S) wavelength-selective cones). As the intensity of light in their preferred wavelength band is increased, the sensitivity of the photoreceptors decreases, cancelling out the change in the illumination spectrum. For example, as the sun sets and the long-wavelength light reflected from the scene increases, the sensitivity of the long-wavelength cones decreases, and therefore their excitation tends to remain the same. The notion that the response of each distinct type of receptor may be continually adjusted by a measure of its average stimulation over the whole scene was first proposed by J. von Kries in 1902, before the receptors themselves had been identified.3. Models of colour constancy
Many theoretical models of colour constancy are based on von Kries-like adaptation of the receptors or the colour-opponent neurons higher in the visual pathway. The question still remains, though, as to how and where — neurally — the necessary sampling of reflected light over the whole scene is performed. It might be that rapid eye movements successively expose the photoreceptors to different parts of the scene, thereby enabling the average to be computed in the retina. Or, it might be that neurons with large receptive fields in the cerebral cortex simultaneously pool information from large regions. In his landmark model, Edwin Land (1964) proposed a specific method for computing the relative reflectance values of a surface in three chromatic channels, by tracking the ratio between the reflected light intensities at nearby points along many random paths starting from the target surface. Land's algorithm was a prescription for a robot, not a description of biology, but in naming it the Retinex, he accorded the retina and cerebral cortex equal probability as sites for similar mechanisms in the human brain. What is known is that chromatic adaptation occurs very quickly, being nearly complete in less than one minute.The hallmark of von Kries-like models is that they could be implemented at very early levels of the visual pathway, before the brain 'parses' the image into its distinct components. Other mechanisms likely to contribute to colour constancy require more sophisticated image processing: for example, specular highlights (mirror-like reflections of incident light from glossy surfaces) carry direct information about the colour of the illumination. But for this information to be used, bright spots in an image must first be identified for what they are. At an even higher level, cognition may intervene: it may be that we recognize and memorize the colour of familiar objects such as our own hands, and, when the illumination changes, use their colour to calibrate the colours of other objects in the scene.
All modern explanations of colour constancy acknowledge that it can only occur for surfaces in the context of other surfaces: the phenomenon requires comparisons between the light reflected from a surface and its surroundings. Thus, in a seemingly paradoxical way, colour constancy may explain the colour mutability that artists have exploited for centuries: colours depend on the colours we have just seen, nearby in either space or time.
4. Simultaneous colour contrast
This phenomenon is perhaps the most striking example of how colours depend on their context. In the late 19th century, Michel Chevreul, director of dyeing at the Gobelin factory in Paris, observed that the difference between two opposing colours is dramatically enhanced when they are juxtaposed: reds appear redder when placed next to green, blues more blue when placed next to orange, darks darker when surrounded by lights. A small square painted with grey ink will appear pink against a background painted green; it appears greenish when surrounded by red. Chevreul meticulously formulated many such laws of colour contrast, prefiguring the computerized colour appearance models used in today's colour industries.Recent experiments show that, for simple configurations, the colour appearance of surface may be simply predicted by the ratio between the cone excitations from it and its background. Since changes in natural daylight tend to preserve this ratio, it may be that encoding colours as cone contrasts evolved as a means of achieving colour constancy (Foster and Nascimento 1994).
5. Colour assimilation
In simultaneous contrast, the surrounding colour induces its opponent colour in the target surface. In the lesser-known but equally vivid phenomenon of colour assimilation, the surrounding colour induces the same colour as itself in the target surface. For example, take a chequerboard pattern of alternating green and purple squares. In the centre of the chequerboard, replace a small number of green squares with grey squares. These grey squares will take on a purplish hue, like their immediate surroundings. If the grey squares instead replace the purple squares, they take on a greenish hue.Why does colour assimilation occur in some patterns, and contrast in others? The full answer is not known, but one rule seems to be: when the surrounding colour is periodically interleaved with the target colour on a fine spatial scale, the colours assimilate. This may be because fine-scale variations in colour are more likely to result from a pattern on a single material than from local changes in illumination; so to estimate any bias induced by the global illumination, these variations must be smoothed, or assimilated. Once assimilated, the common effects of the illumination must be factored out; hence, the need for contrast on a large scale.
Thus, colour contrast and assimilation may both result from a natural striving towards colour constancy in an unnatural situation. The pictures with which we illustrate contextual influences on colour are not like the sunlit vistas in which colour constancy evolved. When the scene does not meet the ancient criteria laid down, constancy mechanisms may fail, with spectacular consequences. Indeed, colour constancy fails under artificial lights (notably, many fluorescent lights) whose energy spectra are not smooth but jagged.
6. Measurements of colour perception
The fact that we can measure the failure of colour constancy does not mean that we can now measure qualia. We can only measure the physical properties of lights and surfaces and record behavioural responses, typically by asking the observer(a) to report whether two distinct light spectra appear the same or different,
(b) to adjust the components of a coloured light to match another, or
(c) to name the colour.
Matching by adjustment is the technique most often used to demonstrate colour constancy. For example, the observer may be asked to adjust the light reflected from a surface so that it appears 'grey' or 'achromatic' under the particular light source illuminating the scene. If the observer adjusts the light to be the same as that reflected by a true 'grey' surface reflectance (say, if he reproduces the yellowish light that a neutral surface reflects under a yellowish source) then he shows perfect colour constancy. Intriguingly, observers often show relatively poor colour constancy when it is measured with simulated papers and light sources displayed on a computer screen. Yet, measurements under more natural conditions, with real surfaces and lights, reveal almost perfect constancy (Kraft and Brainard 1999). The discrepancy may be because, under natural conditions, all the higher-order clues contribute in full, including those from specular highlights and cognitive factors. None the less, low-level mechanisms of chromatic adaptation and contrast probably do the bulk of the work. This conclusion is supported by the fact that other animals, including goldfish, honeybees, and hens, all show colour constancy when tested with surface identification tasks (Neumeyer 1998).
Colour naming is the least reliable method for measuring perception. Over the past half-century, anthropologists and psychologists have searched for proof that colour is a human universal. In 1969, Berlin and Kay famously argued that all languages develop at least two of the following basic colour terms in the following evolutionary order: white/black, red, green/yellow, blue, brown, purple/pink/orange/grey. The evidence for linguistic universality is shaky (Saunders and van Brakel 1997). Yet, while colour-naming systems vary widely across cultures, basic colour discrimination abilities do not: the Quechi Indians have only one word for blue and green, yet put them in different perceptual categories (Davidoff 1991).
Thus, although no one can experience your perception of yellow, or prove that it is the same as mine, colour scientists can predict which physical and physiological conditions are necessary for you to report the yellow you claimed earlier. In the contextual dependence of colours lies a fundamental similarity between minds — the way in which colours vary is the same from person to person. Despite their subjectivity, colours are trustworthy measures of real object properties.
(Published 2004)
— Anya C. Hurlbert
- Bibliography
- Davidoff, J. (1991). Cognition through Colour.
- Foster D. H., and Nascimento, S. M. C. (1994). 'Relational colour constancy from invariant cone-excitation ratios'. Proceedings of the Royal Society of London Series B Biological Sciences, 257.
- Gottlieb, A. (2000). The Dream of Reason.
- Kraft, J. M., and Brainard, D. H. (1999). 'Mechanisms of color constancy under nearly natural viewing'. Proceedings of the National Academy of Sciences of the USA, 96.
- Land, E. H. (1964). 'The retinex'. American Scientist, 52.
- Neumeyer, C. (1998). 'Comparative aspects of colour constancy'. In Walsh, V., and Kulikowski, J. J. (eds.), Perceptual Constancy: Why Things Look as They Do.
- Saunders, B. A. C., and van Brakel, J. (1997). 'Are there nontrivial constraints on colour categorization?' Behavioural and Brain Sciences, 20.
