Abstract
How does information processing differ between upright and inverted face recognition? Previous research from our lab suggests that, in the spatial domain, the differences in processing are quantitative rather than qualitative (Sekuler et al., 2004). Here we ask whether differences exist in the Fourier domain. Previous studies suggest that observers rely on frequency information centered around 9 cyc/face when recognizing upright faces (Nasanen 1999; Gold et al., 1998; Gold et al., 1999). Observers might be worse at inverted face recognition because they rely on a different, or broader range, of spatial frequencies for inverted face recognition than for upright face recognition. A recent study by Nakayama (VSS 2003) suggests that inverted face recognition might be less selective for frequency than upright face recognition, however spatial frequency tuning was not measured directly. We re-examined the issue using critical-band masking to measure spatial frequency tuning in a 10-alternative face recognition task, for both upright and inverted faces. In agreement with past results, upright face recognition relied on a narrow band channel, ∼1.4 octaves, centered at ∼8 cyc/face. However, despite the fact that observers required significantly more contrast to discriminate upside-down faces than upright faces, observers used a similar narrow band of spatial frequencies regardless of face orientation. Because our stimuli and task differed considerably from those of Nakayama, additional research is needed to elucidate the nature of spatial frequency selectivity under various conditions. Regardless, our results place strong constraints on how the strategies for upright and inverted face recognition might differ. We propose that the critical difference lies not in which frequencies are used, but in how information is used within a narrow band of frequencies.
This research was supported by the Canada Research Chair Program and NSERC Discovery Grants #42133 & 105494.