Abstract
Facial identity and facial expression represent invariant and changeable aspects of faces, respectively. The current study investigated how human observers (n=5) use spatial frequency information to recognize identity versus expression. We measured contrast thresholds for the identification of faces with varying expression and for the recognition of facial expressions across varying identity as a function of the center spatial frequency of narrow-band additive spatial noise. At a viewing distance of 60 cm, the peak threshold representing maximum sensitivity was at 11 cycles/face width for identifying the faces of two males or two females with varying expression. The peak threshold was significantly higher for recognizing facial expressions across varying identity: it was at 16 cycles/face width for discriminating between happiness and sadness, and between fear and anger, whether the expression was high or low in intensity. In a second phase we investigated the effect of viewing distance. As viewing distance increased from 60 to 120 and 180 cm, the peak threshold for identifying faces shifted gradually from 11 to 8 cycle/face width, while the peak threshold for recognizing facial expressions shifted gradually from 16 to 11 cycles/face width. The patterns from human observers were different from an ideal observer using all available information, which behaved similarly in recognizing identity and expression. In conclusion, we found, regardless of viewing distance, the optimal spatial frequency band for the recognition of facial expressions is higher than that for the identification of faces. The patterns suggest that finer details are necessary for recognizing facial expressions than for identifying faces and that the system is only partially scale invariant.
Canadian Natural Sciences and Engineering Research Council (NSERC).