Abstract

This paper proposes a new block-matching motion estimation algorithm for a lip reading user authentication system. The method described in this paper represents a sub-system of the Silent Pass project. Silent Pass is a lip reading password entry system and person authentication project for security applications. The aim is to provide a complete solution for secured access to ATMs and internet services in multi-media environments. The proposed block matching algorithm is an optimal algorithm that mimics the actions of an exhaustive full search block matching algorithm (FSBM). The proposed algorithm reduces the computational load of FSBM by pruning non-candidate blocks from the search window using approximate SAD (sum of absolute differences) values. This computational reduction leads to enhanced performance in terms of speed and reduced computational load for motion estimation. Lip reading authentication is a real-time application that makes FSBM an unrealistic choice because of its low speed. The proposed algorithm provides the same result as the FSBM algorithm at a high speed.

Key words: Human computer interaction (HCI), full search block matching (FSBM), pruning full search block matching (PFSBM), VLSI architecture.