Face distributions in similarity space under varying head pose

摘要

Real-time identity-independent estimation of head pose from prototype images is a perplexing task requiring pose-invariant face detection. The problem is exacerbated by changes in illumination, identity and facial position. We approach the problem using a view-based statistical learning technique based on similarity of images to prototypes. For this method to be effective, facial images must be transformed in such a way as to emphasise differences in pose while suppressing differences in identity. We investigate appropriate transformations for use with a similarity-to-prototypes philosophy. The results show that orientation-selective Gabor filters enhance differences in pose and that different filter orientations are optimal at different poses. In contrast, principal component analysis (PCA) was found to provide an identity-invariant representation in which similarities can be calculated more robustly. We also investigate the angular resolution at which pose changes can be resolved using our methods. An angular resolution of 10° was found to be sufficiently discriminable at some poses but not at others, while 20° is quite acceptable at most poses.