COMBINING EVIDENCE IN PERSONAL IDENTITY VERIFICATION SYSTEMS

A methodology for fusing multiple instances of biometric data to impro ve the performance of a personal identity verification system is devel oped. The fusion problem is formulated in the framework of the Bayesia n estimation theory. The effect of different fusion strategies on the error probability is analysed theoretically. The proposed methodology is then demonstrated on the problem of personal identity verification using multiple facial images. Experimental studies on the M2VTS databa se confirm the predicted improvements in performance. A reduction in e rror rates of up to 40% is achieved. The performance gains are initial ly monotonic but they tend to saturate after integrating the first few observations. It is also shown that the fusion based on rank order st atistic, i.e., the median, is robust to outliers. (C) 1997 Elsevier Sc ience B.V.