Face recognition using virtual parametric eigenspace signatures

This paper proposes a novel system for face recognition which utilises the parametric eigenspace of Murase and Nayar (1993) to determine both the pose and identification of a face held within a current face database. In order that all views from profile to frontal are available for recognition a sample of views are taken from a range of angles so that, during the identification process, virtual views can be estimated from a novel view to supply ordered coordinates in the eigenvector subspace for the recognition algorithm. The method is demonstrated on a set of faces and results show that the technique is practical. The system is described in the context of a theoretical model of human face recognition which closely resembles that described in the literature.