Abstract
Recent studies have demonstrated that observers can readily extract the average value of a stimulus set, for dimensions ranging from motion to orientation to facial expressions. We have examined the ability to judge the average value of pairs of colors. Stimuli were composed of two alternating colors displayed in an 11 by 11 array of x deg spots. Observers adjusted one or both colors to a specified average (e.g. gray or a balanced blue-green) or so that the two colors appeared complementary. Performance on these tasks suggest that observers are very good at judging the relative amounts of different colors; while poor at inferring the actual average value of different color combinations, especially for pairs that include different color categories. In current work, we explore the internal “color space” in which observers quantify the colors, by comparing the balance for different color pairs predicted by a cone-opponent vs. perceptually-opponent representation of the components. The latter was derived from hue-scaling of the component colors (e.g. to estimate the perceived proportion of blue or green within each component). These individual proportions fail to predict the average proportions selected for the full set of color pairs, which also deviate from the pattern predicted by averaging the linear or log cone-opponent contrasts. Thus judgments of hue combinations are nonlinear in a way that does not directly reflect simple putative models of early (cardinal axes) or late (Hering axes) color coding.
Supported by EY-10834