Audio-visual Based Person Recognition with Fusion at Feature Level

摘要

This paper proposes an Audio-Visual based Text-Independent (AVTI) person recognition system in which multiple biometrics are integrated at the feature fusion level and the subjects are then classified by analyzing their probability density functions. To this end, a feature synchronization strategy is proposed to fuse the features extracted from the audio and visual signals. Test results from applying the proposed algorithm to a virtual AVTI database show that it achieves a better recognition rate, and is superior to any of the individual biometric systems it derives from.