Re: the color constancy problem

Hello Dan – thanks for the question, behind which lies a world of complexity. I’ll try to answer as simply as possible. You are correct in your understanding of the large changes we can measure in colour according to illumination, and that despite this, our perception of colour is relatively independent of illumination. One of the most dramatic illustrations of this I have seen, demonstrated by Professor R. W. G. (Bob) Hunt, involves a slide (actually three identical slides) of a scene that included a girl wearing a white blouse, beside a grey card and a set of colour patches. Also in the scene was a bright yellow cushion. The first copy of the slide was projected with no filtration. The second had a transparent cyan filter cut to lie on top of the cushion and no other part of the image. When projected, the cushion now became a deep green colour. The third slide used exactly the same filter, but this covered the whole slide, not just the cushion. Amazingly, the cushion was restored to its original yellow hue, albeit relatively unsaturated. So, in slides 2 & 3 we see exactly the same wavelength and brightness of light coming from the cushion area of the projected image, but our perception translates one as green (spectrally true) and the other yellow.

The key here is our perception. The world of our experience is not only coloured, but is also on average neutral when you integrate all the colours in all scenes together. On top of this, our experience teaches us to have some reference points in respect of hue. Flesh tones, whites, greys that we recognise as such are examples. Sometimes this can fool us, especially if the scene only contains a single item of recognition from a hue standpoint. So in the example above, the girl’s flesh, her white blouse, the grey card and to some extent the colour patches act as references which force us to perceive the cushion to be yellow, in order to make the rest of the scene be perceived correctly according to our reference values. In other words, our perception of colour is very much a relative, and not an absolute thing.

So, you ask - how important is this to us? In a sense there is no answer to that. It is an inevitable consequence of our perceptual system, which refers all the time to experience. One might speculate that without that mechanism, if we saw as a spectrophotometer does, we would see differences in colour all the time. Interpretation of that might be a problem and induce nausea or other confusing side-effects. As you can see from the artificial example I described, we can cope with very dramatic spectral shifts. In normal scenes these can also be large – and you only have to look at the difference between colour rendition by photographic film when the same item is taken under daylight and tungsten illumination to see a measure of this. The film make a straight physical record, with no perceptual interpretation. (Special films balanced for tungsten illumination are available to give more pleasing results under taht illumination.) Digital cameras need automatic “white balance” or other means of integrating the scene to neutral to get pleasing results – for the same reason. Tungsten lighting is much lower in temperature than daylight – much less blue, therefore more yellow in the illumination.

Here’s an interesting experiment to do, if you have a digicam, capable of setting a fixed white balance. Set the white balance for daylight and then use it set on that to take an indoor picture under tungsten illumination. Make a print – without any colour adjustment, and view the print in daylight at normal viewing distance. It should look yellow (i.e. display an overall and objectionable yellow cast). View the same image on your computer screen in a brightly lit (preferably by daylight) room. Still got a nasty yellow cast? Then view the same image on your computer screen at night with all other room lighting off. Does it still look yellow? If not, why not?