Abstract
When processing a face stimulus, the human visual system tends to strongly integrate its constituent features (eyes, nose, mouth, etc) in a so-called holistic representation. Such feature integration mainly occurs in face-sensitive regions located in bilateral fusiform gyrii. Behavioural studies showed that feature integration relies on the extraction of low spatial frequencies (LSF) while high SF (HSF) underlie more local aspects of feature analysis.
Following coarse-to-fine models of vision, we propose that the LSF-driven feature integration is an early and fast stage of face perception, in contrast to the longer-lasting extraction of detailed feature cues in HSF.
By means of an event-related fMRI design, the present study investigated the temporal dynamics of face LSF and HSF processing in the network of face-sensitive cortical regions. Faces were flashed at 75, 150, or 300 msec, followed by a Gaussian mask. They were band-pass filtered to preserve low or high SF. At short stimulus durations, face-sensitive regions located in bilateral fusiform gyrii and superior temporal sulci responded more strongly to LSF than HSF faces. At longer durations, the same regions were more active for HSF than LSF faces. This pattern did not replicate for phase-scrambled versions of the stimuli.
Taken together our findings suggest that face perception proceeds following a coarse-to-fine scenario, with an early and fast LSF-driven feature integration being relayed by the slower accumulation of HSF local information.
We thank Julie Haubrechts for her help during the recording and processing of fMRI data.