Visual Perception

Reflections on colour constancy

Article metrics

A study of colour perception shows that, when assigning colour to objects, the seeing brain takes into account subtle reflections of light between the surfaces in a scene.

For more than two centuries, scientists and artists have come up with a range of ways to demonstrate that the wavelength composition over a whole scene can affect how we perceive the colour of the individual parts of that scene. The proportion of light of each wavelength reflected from an object can be highly beneficial in detecting or recognizing objects1. However, to make use of that invariant, the visual system somehow has to discount the illuminating light, which can vary quite drastically.

This process is usually called ‘colour constancy’, and the degree to which it is shown depends on many factors. Some of them occur at the early stages of sensory processing2, such as local colour contrast, whereas others (for instance, colour memory) occur at higher cognitive levels3. Most computational schemes for achieving colour constancy try to decompose the overall light reaching the eye into one component that is due to the illuminant, and a second component due to the reflectance.

However, the physics of light is more complicated than the simple reflection of light from a surface into the eye, as if looking at a photograph. In a three-dimensional world, some light is reflected from one surface, but it then bounces to yet another surface from which it is reflected into the eye (Fig. 1). And so on. These indirect reflections are called ‘inter-reflections’, and are of especial interest to those involved in computer graphics and computer vision4. For example, computer simulations of indoor scenes appear more realistic when inter-reflections are taken into account. Indeed, many recent advances in computer graphics are due to the discovery of efficient algorithms to calculate the effects of all such multiple bounces of light among the vast number of surfaces typically contained in a scene.

Figure 1: Illustration of inter-reflections.
figure1

Part of the light is being scattered directly into the eye of the observer (green line). Other light rays enter the eye after being reflected from the magenta side of the card onto the white side (grey line). The magenta is ‘bleeding’ into the white.

Reporting on page 877 of this issue, Bloj et al.5 tested whether inter-reflections are taken into account by the human visual system when perceiving the colour of surfaces. To do this, they presented subjects with a folded card, with its two sides opening towards the observer (Fig. 2). One side of the card was painted magenta, the other white. The magenta half reflects light only in part of the spectrum, some of which reaches the eye directly, and some of which is reflected onto the white side of the card. As a result, the light reflected from the white side is no longer balanced over all wavelengths, and it should appear a quite saturated magenta under the conditions of the experiment. Surprisingly, subjects reported that it appeared only slightly pinkish, suggesting that the visual system discounts the effect of inter-reflections to compute surface reflectance. In other words, the colour we perceive is influenced not only by the two-dimensional image of an object projected onto the retina, but also by our perception of the object's three-dimensional shape.

Figure 2: The experiment done by Bloj et al.5.
figure2

Observers viewed a folded card opening towards them. a, Under normal viewing, observers discount the effect of inter-reflections. b, When seen through a pseudoscope, the inter-reflections are interpreted as a change in surface colour because they are incompatible with the percept of the bent-away card. The magnitude of the effect has been emphasized for clarity.

The strength of Bloj and colleagues' paper lies in the way they controlled for potential confounding factors. They used a special device called a pseudoscope, which optically inverts the depths of all objects in the scene. This means that the card will appear to open away from the observer when seen through the pseudoscope (Fig. 2b). This leads to an identical two-dimensional distribution of image intensities over the visual field. Perceptually, however, there is no longer the possibility for inter-reflections between the two surfaces, because they no longer seem to be pointing towards one another.

In this way, the physical stimulus was unchanged, so the inter-reflections and their effect were still present. But they could no longer be interpreted as such by the brain. Instead, the bias in the wavelength distribution of the card seemed to be due to its surface reflectance, and, as a result, it was perceived to have a deep pinkish hue. So the results show not only that three-dimensional shape can affect colour perception, but also that the seeing brain indeed knows the physics of light inter-reflections.

Similar phenomena are well known in our perception of brightness. When dealing with black and white scenes, the effects of shape, reflectance and illumination are basically interchangeable, presenting the visual system with an infinite number of possibilities for interpreting such scenes6. Yet we perceive the world as being quite stable. The solution seems to be that the most likely percept is calculated, so changes in the perception of three-dimensional shape can easily (and profoundly) affect our perception of brightness7.

The question is, then, how much knowledge of the physics of light has the visual system acquired? It seems to know even the very subtle aspects, which may have only a very limited function when viewing natural outdoor scenes. For example, the geometric information contained in highlights — generated by mirror-like reflections from glossy surfaces — can affect the perception of surface curvature8. In their experimental set-up, Bloj et al.5 chose conditions that emphasized the role of inter-reflections. The white side of the card almost lay in the shadow, so it received very little direct light; most light came from the magenta card. In typical real-life situations, however, a much smaller part of the light reaching the eye is due to inter-reflections.

A tremendous amount of psychophysical characterization has been done on visual ‘illusions’, such as those mentioned above, yet we know hardly anything about the brain processes that underlie such phenomena. The analysis of colour and other elementary features such as motion or orientation is often assumed to occur at an early stage of sensory processing in the visual system, whereas information about surfaces and objects is thought to be extracted during the later stages. Bloj and colleagues' results are difficult to reconcile with this simplified idea. One has to differentiate between the different uses of colour information in visual perception. Colour might be used at an early stage to segment objects from one another and from the background, but at that stage there is no need to assign a definitive colour to each object. Higher visual areas (such as the infero-temporal cortex, which is known to be important for object recognition9) also show a large degree of colour selectivity10, and Bloj and colleagues' results support the idea that the colours we perceive may be determined at a rather late stage of visual processing.

References

  1. 1

    Mollon, J. D. & Jordan, G. J. Exp. Biol. 146, 21–38 (1988).

  2. 2

    Kraft, J. M. & Brainard, D. H. Proc. Natl Acad. Sci. USA 96, 307–312 (1999).

  3. 3

    Troost, J. M. & de Weert, C. M. Percept. Psychophys. 50, 591–602 (1991).

  4. 4

    Foley, J. D., van Dam, A., Feiner, S. K. & Hughes, J. F. Computer Graphics: Principles and Practice (Addison–Wesley, Reading, Massachusetts, 1990).

  5. 5

    Bloj, M. G., Kersten, D. & Hurlbert, A. C. Nature 402, 877–879 (1999).

  6. 6

    Adelson, E. H. & Pentland, A. P. in Perception as Bayesian Inference (eds Knill, D. C. & Richards, W.) 409–423 (Cambridge Univ. Press, New York, 1996).

  7. 7

    Knill, D. C. & Kersten, D. Nature 351, 228–230 (1991).

  8. 8

    Blake, A. & Bülthoff, H. H. Nature 343, 165–168 (1990).

  9. 9

    Logothetis, N. K. & Sheinberg, D. L. Annu. Rev. Neurosci. 19, 577–621 (1996).

  10. 10

    Komatsu, H., Ideura, Y., Kaji, S. & Yamane, S. J. Neurosci. 12, 408–424 (1992).

Download references

Author information

Correspondence to Karl R. Gegenfurtner.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Gegenfurtner, K. Reflections on colour constancy. Nature 402, 855–856 (1999) doi:10.1038/47194

Download citation

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.