Multi-sensory pattern analysis for person identification with synergetic computers

Nowadays automatic person identification becomes more and more important. With increasing computing power available, it is possible to solve person identification problems using two different input sources. In an earlier paper we have already used combined optical lip reading and acoustical voice analysis for identifying the person speaking. The classification of the preprocessed data is done separately by synergetic algorithms, which today attract increasing attention as quick and robust algorithms for solving industrial classification tasks. Since synergetic algorithms, also called synergetic computers, show a close mathematical similarity to self-organizing phenomena in nature, they present a clear perspective for hardware realizations. In this paper we present the results of an improved preprocessing of the acoustical data of the person identification system presented in "T. Wagner, T. Tollen, F. Boebel:'Multi-Sensorial Inputs for Opto-Acoustic Identification with Synergetic Comput ers, Proceedings of the International Conference on Artificial Neural Networks, Sorrento, pp. 1392-1395, 1994". The improved algorithm was tested on the same data obtained from the field test containing 2020 pattern from 101 different persons. There was again no misclassification, but a much lower rejection rate of only 8 per cent