Face distributions in similarity space under varying head pose

Real-time identity-independent estimation of head pose from prototype image s is a perplexing task requiring pose-invariant face detection. The problem is exacerbated by changes in illumination, identity and facial position. W e approach the problem using a view-based statistical learning technique ba sed on similarity of images to prototypes. For this method to be effective, facial images must be transformed in such a way as to emphasise difference s in pose while suppressing differences in identity. We investigate appropr iate transformations for use with a similarity-to-prototypes philosophy. Th e results show that orientation-selective Gabor filters enhance differences in pose and that different filter orientations are optimal at different po ses. In contrast, principal component analysis (PCA) was found to provide a n identity-invariant representation in which similarities can be calculated more robustly. We also investigate the angular resolution at which pose ch anges can be resolved using our methods. An angular resolution of 10 degree s was found to be sufficiently discriminable at some poses but not at other s, while 20 degrees is quite acceptable at most poses. (C) 2001 Elsevier Sc ience B.V. All rights reserved.