Recognition of unfamiliar faces

Abstract

People are excellent at identifying faces familiar to them, even from very low quality images, but are bad at recognising, or even matching, faces that are unfamiliar. In this review we shall consider some of the factors which affect our abilities to match unfamiliar faces. Major differences in orientation (e.g. inversion) or greyscale information (e.g. negation) affect face processing dramatically, and such effects are informative about the nature of the representations derived from unfamiliar faces, suggesting that these are based on relatively low-level image descriptions. Consistent with this, even relatively minor differences in lighting and viewpoint create problems for human face matching, leading to potentially important problems over the use of images from security video images. The relationships between different parts of the face (its "configuration") are as important to the impression created of an upright face as local features themselves, suggesting further constraints on the representations derived from faces. The review then turns to consider what computer face recognition systems may contribute to understanding both the theory and the practical problems of face identification. Computer systems can be used as an aid to person identification, but also in an attempt to model human perceptual processes. There are many approaches to computer recognition of faces, including ones based on low-level image analysis of whole face images, which have potential as models of human performance. Some systems show significant correlations with human perceptions of the same faces, for example recognising distinctive faces more easily. In some circumstances, some systems may exceed human abilities on unfamiliar faces. Finally, we look to the future of work in this area, that will incorporate motion and three-dimensional shape information.