Scene statistics and chromatic induction: only the local correlation between luminance and chromaticity matters

Abstract

Colour constancy is the ability to identify objects' colours fairly consistently, despite considerable changes in the spectral composition of the illuminant. This ability involves disentangling the influences of illumination and reflectance. Golz & Macleod (2002) suggested that the correlation between luminance and colour within a scene can help to do so. They pointed out that the brightest surfaces will be the ones that reflect well in the colour of the illuminant. In line with this idea, they showed that chromatic induction is not only a function of the mean luminance and chromaticity in a scene, but also depends on the distribution of colours across the brighter and darker surfaces. We examined whether the correlation between luminance and colour is determined for the whole scene. We compared scenes with the same average luminance and chromaticity, but with different correlations between luminance and chromaticity. Our results confirm that there is a shift in perceived colour (chromatic induction) away from the chromaticity of bright surfaces. By comparing several spatial configurations we show that only the correlation within about 1° of the target matters. Thus, the proposed role of the correlation between luminance and colour in maintaining colour constancy is presumably incorporated within the way in which the colour contrast at borders is analysed.