Robust User Identification Based on Facial Action Units Unaffected by Users' Emotions

Abstract

We report on promising results concerning the identification of a user just based on its facial action units. The related Random Forests classifier which analyzed facial action unit activity captured by an ordinary webcam achieved very good values for accuracy (97.24 percent) and specificity (99.92 percent). In combination with a PIN request the degree of specificity raised to over 99.999 percent. The proposed biometrical method is unaffected by a user's emotions, easy to use, cost efficient, non-invasive, and contact-free and can be used in human-machine interaction as well as in secure access control systems.