Testing principal component representations for faces

Citation

Hancock PJB, Bruce V & Burton AM (1998) Testing principal component representations for faces. In: Bullinaria J, Glasspool D & Houghton G (eds.) 4th Neural Computation and Psychology Workshop, London, 9–11 April 1997: Connectionist Representations. Perspectives in Neural Computing. 4th Neural Computation and Psychology Workshop, London, 09.04.1997-11.04.1997. London: Springer, pp. 84-97. http://link.springer.com/chapter/10.1007/978-1-4471-1546-5_8; https://doi.org/10.1007/978-1-4471-1546-5_8

Abstract
A variety of experimental results indicate that the human visual system processes faces at least to some extent holistically, rather than by analysing individual features such as nose and eyes. Principal Components Analysis (PCA) of face images, which is widely used in engineering approaches to face identification, produces an inherently global representation. We investigate the psychological plausibility of this representation, looking at correlations with human perceptions of memorability and similarity. We show that transformation of faces to an average shape prior to PCA improves correlations with human ratings

People

Professor Peter Hancock

Professor, Psychology