Abstract
Recent studies indicate that facial motion cues benefit recognition more for familiar faces than unfamiliar faces. This performance gap underscores a fundamental question: How much and what kind of familiarity with a moving face is needed to improve recognition? Using a familiarization paradigm, we tested participants' recognition performance for newly-learned faces from novel test formats. In Exp. 1, participants watched whole-body videos of people walking toward the camera (“gait videos”). Participants viewed the videos either once, twice, or four times and were asked at test to recognize these people from static pictures or from videos of faces speaking. In Exp. 2, participants learned pictures or videos of faces and were tested with gait videos. In both experiments, the results showed that a modest amount of face or person familiarization, in the form of pure repetition, was sufficient to improve performance across large changes in image/video format. In Exp. 2, motion was helpful only in the 4-exposure condition. An analysis combining the results of the two experiments showed that the transfer from face to gait was more accurate than the transfer from gait to face. Further, “motion-motion” transfers (between speaking-faces and gait videos) were more accurate than “static-motion” transfers (between static faces and gait videos). In Exp. 3, participants learned pictures or videos of frontal-view faces and were tested with static profiles. We found no motion advantage. This result indicates that structure-from-motion processes do not contribute to the motion advantage found in Exp. 2. We discuss these results in terms of the “dual-route” hypothesis, by which the processing of the moving and static aspects of faces and people may proceed independently.
Supported by grants and contracts from ONR/DARPA and TSWG awarded to A.O'T. and H.A.