Eigenphase-based face recognition: a comparison of phase-information extraction methods

Phase information has recently been used quite frequently in the field of face recognition, especially when the facial images are taken under variable illumination conditions. The approach which combines a PCA with the phase information in order to extract the facial features and reduce the dimensionality of the feature space is called eigenphase method. This paper focuses on four modifications to the eigenphase method. The methods differ in terms of their approach to obtaining the phase information from the facial images. The modifications are evaluated on the XM2VTS, Yale and ORL datasets in order to examine their robustness, primarily under variable and normalized illumination conditions and for slightly variable head poses and facial expressions. The results of the experiments are presented and discussed.