Abstract
Studies of texture perception and ensemble coding have shown that human vision is highly sensitive to differences between color distributions’ simple statistics, such as their mean chromaticity or variance. We asked how sensitive observers are to the density of distinct colors defining a texture. Stimuli were 10×10 square palettes, with the color of each element chosen pseudo-randomly from sets of 2 to 6. Specific colors varied randomly on each trial but were defined in a scaled MacLeod-Boynton space to be 180 deg apart for a setsize of 2, 120 deg for a setsize of 3, etc. In behavioral experiments, 4 palettes were shown and subjects picked which contained an extra color. Conditions varied as to whether the 3 common palettes had the same colors or the same setsize but random colors. For both, as setsize increased from 2 to 3 or 4, accuracy for brief presentations fell to chance and response times for continuous presentations greatly increased. This was despite maintained sensitivity for control conditions where the target had the same setsize but rotated by a comparable hue angle. In a second experiment we assessed performance using an EEG paradigm where images with a fixed setsize were shown at 6 Hz and a target ensemble was embedded at 1 Hz. Only the 2 vs. 3 setsize difference yielded a significant 1 Hz response. Our results are similar to the grayscale texture metamers of Chubb et al. (1993) in showing very limited preattentive sensitivity to the number of colors in an ensemble.