Camile A. Franke¹ (), Jesse R. Macyczko¹, Michael A. Webster¹; ¹University of Nevada, Reno

Variations in facial color have been associated with differences in both biological sex and facial expression. We examined priors for the chromaticity of different face categories by measuring the chromatic range that appeared natural. Stimuli were images from the Chicago Face Database and included 3 females and 3 males (white race) modeling neutral, angry, or happy expressions. The faces were cropped to remove external features and adjusted so that all had the same mean chromaticity (for which ground truth was not known). The mean chromaticity of each face was then further varied within a version of DKL color space by rotating the hue angle (+/- 32 deg in 4 deg steps) or scaling the saturation (0 to 3x the original chromatic contrast in 0.2 steps), forming 33 stimuli for each individual and expression. These were shown in random order on a display with a visual angle width of 10 deg, remaining on screen until participants gave their responses. Ten observers (7 female, 3 male) rated whether each face color appeared “natural” or “unnatural,” with the mean of the natural gamut estimated for each face and observer. Results were assessed with separate 2-way ANOVAs (sex by expression) for hue and saturation. Hue gamuts showed weak but significant effects for both factors: averages for male and (surprisingly) happy faces were more red. Male gamuts also had significantly higher saturation than female, with no effect of expression. Our results suggest that observers have small but significant differences in the priors for chromaticity for different categories of faces.