Abstract
Whole objects, like faces, can strongly attract visual attention. However, little is known about how spatial attention is distributed across the face. During 10AFC identification, we presented a small probe pattern at random locations 106ms after each target-face display. Probe patterns were small squares consisting of high-contrast black and white strips. On the premise that detection speed for probes is enhanced by local attention on the face, probe reaction times should reveal how attention is distributed across the face, in both space and time. We used 3 target-face durations. Face identification accuracy, above-chance for the lowest duration, improved with duration for all 4 subjects. The main data are maps of speed, 1/log(RT), across 3 face durations, for 4 subjects. Each map is based on 60 locations, arranged in an elliptical fashion around fixation (the average center of contrast energy across all faces). Using bootstrap percentile tests, we identified probe locations whose speed was significantly above the spatio-temporal average; these locations show us where attention is. At 12ms, attention is weak and spatially uniform; while at 106ms and 200ms, attention is biased to bottom and right parts of the face from the viewers’ perspective. Our results seem surprising given that eye movements (Butler et al, 2005; Peterson & Eckstein, 2012) and classification images (Sekuler et al, 2004) suggest biases toward the top and left. However, the distribution of attention across the face does not necessarily have to match the distribution of extracted information, or fixations. For example, attention might be required to use weak information, less visible regions of the face, or to plan the next fixation. We are currently exploring these possibilities.
Meeting abstract presented at VSS 2015