Audio-Visual Authentication System over the Internet Protocol

In this paper, an audio-visual (AV) authentication system is developed with the objective to increase the robustness of the system to face illumination variation. A multiband feature fusion approach is proposed to search for the mid- and high-frequency subbands that are insensitive to variation in illumination based on wavelet packet decomposition to solve the illumination problem. Simulation results show that the multiband feature fusion approach achieved higher recognition accuracy as compared to a previous study. To further improve the robustness of multiband feature fusion approach to not only invariant to face illumination but also invariant to facial expression variation, principle component analysis is employed to work in conjunction with the multiband feature fusion approach. Then the AV authentication system is implemented over internet protocol (IP) to enable long distance access. In this system, we are concerned about video and audio streaming. Hence, the effects of speech and face compression on recognition performance of AV authentication system over the internet protocol are investigated. The experiment results show that the AV authentication system over IP with smaller data size achieved the same recognition rate as in the standalone system.