Biometric systems are important and practical for the recognition or verification of identities, via such features as faces, fingerprints, and retina patterns. The application of synthetic imagery is one of the latest techniques used to test such systems. This paper reports on the results of two experiments: a pose experiment, in which the angles of the faces are varied, and a temporal experiment, in which the age gaps between the synthetic and reference images are varied.
The results of the temporal experiment indicate that synthetic imagery should be very useful for testing face recognition. For example, it would be fairly difficult to collect a large number of real faces, over an extended period of time, to test the correctness of a biometric system with respect to temporal variations.
On the other hand, the results of the pose experiment “differ significantly” from those of the facial recognition vendor test (FRVT) program conducted in 2000. This is especially important because the FRVT experiments in 2000 were conducted using real images, whereas the present experiment makes use of synthetic images. It is therefore natural for readers to question the validity of synthetic imagery in face recognition testing. Further experiments, or further improvements to the technique, are necessary.
In any case, researchers will benefit from reviewing the results of the experiments, in order to refine their own synthetic imagery techniques.