Audio visual system for large scale people authentication and recognition over internet protocol (IP)

Biometrics is a promising and viable solution to enhance information security systems compared to passwords. However, there are still several issues regarding large-scale deployment of biometrics in realworld situations that need to be resolved before biometrics can be incorporated together. One of these issues is the occurrence of high training time while enrolling a large amount of people into the system. Hence, in this chapter, the authors present the training architecture for an audio visual system for large scale people recognition over internet protocol. In the proposed architecture, a selection criteria divider unit is used to decompose the large scale people or population into smaller groups whereby each group is trained subsequently. As the input dimensions of each group is reduced compared to the original data size, the proposed structure greatly reduces the overall training time required. To combine the scores from all groups, a two-level fusion based on weighted sum rule and max rule is also proposed in this chapter. The implementation results of the proposed system show a great reduction in training time compared to a similar system trained by conventional means without any compromise on the performance of the system. In addition to the proposal of a scalable training architecture for large-scale people recognition based on audio visual data, a literature review of available audio visual speaker recognition systems and large-scale population training architectures are also presented in this chapter.