Abstract
Most researchers now believe that human observers use a default coarse-to-fine strategy (i.e., from low to high spatial frequencies) to extract face information (e.g., Morrisson & Schyns, in press). Schyns, Bonnar & Gosselin (in press) recently discovered an analoguous differential use of face information along the spatial location and spatial frequency dimensions (see also Gosselin & Schyns, 2001). In sum, we know how time and frequency, as well as how frequency and location interact in face recognition; however, we do not know how location and time interact (neither do we know how spatial location, spatial frequency, and time interact, but this is another story). Here, we explore the spatio-temporal use of information in face recognition. Our stimuli set comprised 30 faces (i.e., [5 males + 5 females] * 3 expressions). The stimuli subtended 5.72 × 5.72 deg of visual angle and were presented for 320 ms. We utilized a novel technique called Bubbles (Gosselin & Schyns, 2001) to reveal directly the effective use of visual information. In a nutshell, we sampled space and time with small Gaussian windows (standard deviation = .22 deg in space and 43 ms in time), and adjusted their number on-line to maintain performance at 75% correct. We ran 10 subjects. A proportion-correct-when-available statistics was computed for each pixel (i.e., first order statistics); higher-order statistics were also computed. We obtained clear spatio-temporal modulations of effective use of information.