Abstract
Assessed with briefly exposured Mondrians, color constancy has been reported to vary across a vast range, from almost none, using direct matching of hue and saturation (Arend & Reeves, 1986) to almost complete, when discriminating illuminant from material changes (Craven & Foster, 1992). Not just the task, but the stimuli, labs, conditions, and observers differed between these studies. To isolate any task effect, our naïve observers rated how well the central squares in pairs of successive 1 sec Mondrians matched in material or in hue and saturation. The simulated Mondrians comprised 49 (7×7) abutting surfaces, whose global illuminants were 4000K or 16000K (first Mondrian) and 6700K (second). For observers who clearly distinguished between the two rating tasks, our Brunswickian color constancy index averaged 82% for material and 29% for hue/saturation. For those who did not, the index was ∼ 52%. Inserting fine spatial (5′ arc between squares in a pattern) and temporal gaps (0.2 sec between the first and second pattern) drove the index to ∼ 38% in all cases. If color constancy requires averaging of signals over the display, such averaging is sensitive both to bottom-up signals (physical contiguity) and to top-down ones (task and ability).
Supported by an EPSRC grant to Foster and Reeves.