Automatic Configuration for a Biometrics-Based Physical Access Control System
Selecting appropriate thresholds and fusion rules for a system involving multiple biometric verifiers requires knowledge of the match score statistics for each verifier. While this statistical information can often be measured from training data, that data may not be representative of the environment into which each verifier is deployed. To compensate for missing statistics, we present a technique for estimating the error rates of each verifier using decisions made after a system has been deployed. While this post-deployment data lacks class labels, it is guaranteed to be representative. Extracted error rates can be used to select appropriate fusion rules and search for thresholds that meet operational requirements.
KeywordsInformation System Error Rate Pattern Recognition Data Processing Training Data
Unable to display preview. Download preview PDF.
- 1.Beattie, M., Kumar, B.V.K., Lucey, S., Tonguz, O.: Combining verification deci¬sions in a multi-vendor environment. In: Audio- and Video-based Biometric Person Authentication, AVBPA (2005)Google Scholar
- 5.Phillips, P., Plynn, P.J., Scruggs, T., Bowyer, K.W., Hoffman, J.C., Marques, K., Min, J., Worek, J.: Overview of the face recognition grand challenge. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR (2005)Google Scholar
- 8.Mak, M., Zhang, W., He, M.: Determination of a priori decision threshold for phrase-prompted speaker verification. In: International Workshop on Multimedia Data Storage, Retrieval, Integration and Applications, pp. 96–103 (2000)Google Scholar
- 9.Pierot, J.B., Lindberg, J., Koowaaij, J., Hutter, H.P., Genoud, D., Blomberg, M., Bimbot, F.: A comparison of a priori threshold setting procedures for speaker verification in the cave project. In: International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 1, pp. 125–128 (1998)Google Scholar