Ian Craw, Nicholas Costen, Takashi Kato, Graham Roberts
Automatic Face Recognition: Combining Configuration and Texture
Abstract:We describe in detail a baseline set of procedures for face recognition using Principal Component Analysis. Results are obtained for cues faces in 13 different conditions, with varying lighting and interval between gallery and cue acquisition. We suggest potential improvements to the coding and show how they may be investigated using the baseline setup. Tests show that paying detailed attention to the configuration of faces, and separating the shape and texture into separate vectors significantly improves recognition and allows the modeling of human face recognition phenomena. A theoretical account in terms of a manifold model of facial variation is proposed.