Comparison of racial effect in face identification systems based on Eigenface and GaborJet

Most people report finding it difficult to recognize the faces of people of other races, a phenomenon known as other-race effect, in face perception. If this tendency is also shown by automatic face identification systems, it might pose a type of racial problem for systems that are being widely introduced to security applications on a global scale, such as use for passport control at international airports. Armed with an awareness of this potential problem faced by biometric authentication systems, we investigated whether two major classical schemes for image representation in face recognition, i.e. Eigenface and GaborJet, show different racial effects in terms of verification error on alien subjects. Results indicate that, contrary to general understanding, Eigenface may outperform GaborJet when faces could be either Japanese or non-Japanese.