Dempster-Shafer Theory for Fusing Face Morphing Detectors
Andrey Makrushin, Christian Kraetzer, Jana Dittmann, Clemens Seibold, Anna Hilsmann, Peter Eisert · 2019
Revealing that a human face on a biometric image is a mixture of two or more faces is of immense importance for document issuing authorities and document checking services. If not done, several persons can use the same photo-ID document for identity verification without being condemned. The development of automated face morphing detectors is currently in its early phase. The detectors reported so far are not mature for the market which is reflected in high error rates when tested with “unseen” data. Here, we demonstrate that fusion of several by far non-optimal detectors may lead to significant improvement of detection accuracy compared to that of individual detectors. Among the examined fusion approaches, Dempster's Rule of combination has the best accuracy allowing for coherent decision making even with contradicting decisions of individual detectors.