Abstract
We studied the contribution of color to object and face perception by testing memory colors. Twenty participants (10 female) were asked to match colors of real objects, including real human models, under two conditions: white light, and monochromatic sodium light (589 nm) that renders vision objectively achromatic (Boynton & Purl, 1989). Asymmetric color matches (CIE-Lab) were made by adjusting a color-patch on a monitor (visible through an aperture). Prior work on memory color suggests that under objective achromatic conditions, color judgments are biased toward the typical color of objects. Instead, we found that color judgments for most objects reflected only the monochromatic (yellow) bias of the sodium lamp (mean hue angle of non-face object matches, H = 68°; mean error (match – spectral measure) = 2.0°), and did not differ for objects with diagnostic colors (strawberry, orange, tomato) compared to those without (legos, toy phone, ping-pong ball) (p = 0.2). The exception was faces, which appeared green under sodium light (H = 111°; error = 40.3°; p = 6x10-47) for the two races tested (Caucasian, African-American). The effect was abolished when the face context was masked (error = -1.4°; p = 5x10-3), suggesting a systematic interaction of face-shape and color processing. These results provide compelling evidence for perceptual memory color, but they are paradoxical. The memory color was not normal face color, instead it matched unhealthy complexion. Hierarchical clustering analysis revealed that only the recently evolved L-M color channel, not the older S-channel, mediated the paradoxical memory color for faces. This supports the idea that trichromatic vision was selected in primates for its contribution to social communication by encoding chromatic states of the facial skin (Changizi et al., 2006). The paradoxical nature of the memory color reveals a strong prior for face color, triggering an "unnaturalness" error signal when violated by otherwise reliable visual information.
Meeting abstract presented at VSS 2018