False alarm estimation in an automated fingerprint identification system (AFIS)

Abstract

Law enforcement uses automated fingerprint identification systems (AFIS) to identify people who are reported missing, wanted, or who require background investigation. AFIS employ automated searches of fingerprint features in feature databases. The feature records have relational links to other descriptive information. Quantitative AFIS evaluation includes measurement of missed detection (Type I) and false alarm (Type II) error rates. Some contemporary systems are designed to achieve a very low false alarm rate, creating an environment with sparse data for the error rate estimation. In this environment, expensive large scale testing is needed to observe a few false alarms and error rate estimation is challenging. The authors developed a method for estimating false alarms based on measuring the outcome of data entry processes. The method applies to one-to-many and one-to-one biometric systems used in law enforcement and security. The false alarm rate estimation method can be used during standard benchmark testing. It applies easily to AFIS because the data entry processes check input data to avoid duplicate entries. This authors present the method within the context of AFIS benchmarking. A short tutorial background on AFIS error rate measurement is provided. The theoretical foundation for the method is presented. The formulae used to estimate false alarm rate are presented with the needed extensions for applying them to data entry.