Influence of scene statistics on color constancy

Jürgen Golz* & Donald I. A. MacLeod'
*Christian-Albrechts-Universität zu Kiel, Institut für Psychologie
'University of California, San Diego, Department for Psychology

When the illumination changes from white to chromatic, statistics other than the mean of the chromaticity in a scene can change in a characteristic way. In particular, the correlation between chromaticity and luminance could provide a signed cue, which enables the visual system to evaluate whether an imbalance in chromaticity is due to a chromatic illumination or to a predominance of surfaces with this chromaticity. If, for example, the illumination gets more reddish, the reddish patches in a scene get lighter relative to the other colors; this results in a positive correlation between redness and luminance within the scene. We tested whether this parameters and other high order scene statistics have an influence on color constancy in complex displays.

We used surrounds consisting of a random spatial structure of overlapping circles of a fixed diameter (Mausfeld and Andres, ECVP 1998). For a given condition, the chromaticity and luminance values for the surrounding circles were chosen to achieve a certain high order scene statistic value. Subjects set a center test spot to achromatic gray.

We found a small effect for correlation between chromaticity and luminance. For example, if the surround had a positive correlation between redness and luminance, a more reddish chromaticity was required to make the test field subjectively achromatic. This is the result expected if the observer infers a more reddish illumination in the case of higher correlation values.

The existence of this effect of high order statistics, and its small size, are both consistent with simulations that we have run using natural images: the simulations show that these statistics are of very limited value in diagnosing the illuminant color for natural scenes.