Using Biometric Verification to Estimate Identification Performance

We show how to estimate identification performance for arbitrary gallery size from an ROC curve. The method presented can: (1) estimate an alarm curve; (2) give rank 1 dependence on gallery size; and (3) estimate a CMC curve. We show that the method assumes consistently normalized scores, with normalization the degree of freedom defining the connection between identification and verification. A measure to quantify normalization consistency is defined. Examples are given using both fingerprint and face databases. Further examples are based on the results given in the FRVT 2002 evaluation report.