Evolving faces from principal components

Citation

Hancock PJB (2000) Evolving faces from principal components. Behavior Research Methods, Instruments and Computers, 32 (2), pp. 327-333. https://doi.org/10.3758/BF03207802

Abstract
A system that uses an underlying genetic algorithm to evolve faces in response to user selection is described. The descriptions of faces used by the system are derived from a statistical analysis of a set of faces. The faces used for generation are transformed to an average shape by defining locations around each face and morphing. The shape-free images and shape vectors are then separately subjected to principal components analysis. Novel faces are generated by recombining the image components ("eigenfaces") and then morphing their shape according to the principal components of the shape vectors ("eigenshapes"). The prototype system indicates that such statistical analysis of a set of faces can produce plausible, randomly generated photographic images.

Keywords
Face PCA; facial composites; Evolutionary algorithm; Eigenface; Eigenshape; Face perception; Face physiology; Morphology

Journal
Behavior Research Methods, Instruments and Computers: Volume 32, Issue 2

People

Professor Peter Hancock

Professor, Psychology