Philippe G. Schyns, Harold Hill
Viewpoint Dependence in Face Recognition
Abstract:Face recognition has attracted the attention of vision researchers for the observation
that although most faces are very much alike, people discriminate between them very well.
Even though little is known about the mechanisms of face recognition, a recurrent
phenomenon of general object recognition (viewpoint dependency) could illuminate the way
faces are recognized. In this paper, we investigate in detail the conditions for viewpoint-dependent
face recognition. The first experiment tested whether a particular view of a face
was better and more reliably recognized than other views. Results indicate that when all
views were shown (randomly or in an animated sequence), all were equally well recognized.
The second experiment tested generalization from single views of a face. Different learning
views were found to produce different patterns of generalization. For full-face, performance
fell off with increasing angle of rotation, while for three-quarter there was a peak for the
opposite three-quarter. Profile performances dropped off steeply and there was no recovery
for the opposite profile. Results are discussed in the context of recent psychological,
computational and physiological accounts of viewpoint-dependent object recognition.