INFORMATION AND VIEWPOINT DEPENDENCE IN FACE RECOGNITION

How we recognize faces despite rotations in depth is of great interest to psychologists, computer scientists and neurophysiologists because of the accuracy of human performance despite the intrinsic difficulty of the task. Three experiments are reported here which used three-dime nsional facial surface representations to investigate the effects of r otations in depth on a face recognition task. Experiment 1, using ''sh ape only'' representations, showed that all the views used (full-face, three-quarter and profile) were equally well recognized when all had been learned. Performance was better when the same views were presente d in an animated sequence, rather than at random, suggesting that stru cture-from-motion provides useful information for recognition. When st imuli were presented inverted, performance was worse and there were di fferences in the recognizability of views, demonstrating that the fami liarity of upright faces affects generalization across views. Experime nts 2 and 3 investigated generalization from single views and found pe rformance to be dependent on learned view. In both experiments, genera lization from learned full-face fell off with increasing angle of rota tion. With shape only stimuli, three-quarter views generalized well to each other, even when inverted, but for profiles generalization was e qually bad to all unlearned views. This difference may be explained be cause of the particular relationship of the profile to the axis of sym metry. In Experiment 3, addition of information about superficial prop erties including color and texture facilitated performance, but patter ns of generalization remained substantially the same, emphasizing the importance of underlying shape information. However, generalization fr om the three-quarter view became viewpoint invariant and there was som e evidence for better generalization between profiles. The results are interpreted as showing that three-dimensional shape information is fu ndamental for recognition across rotations in depth, although superfic ial information may also be used to reduce viewpoint dependence. (C) 1 997 Elsevier Science B.V.