When to Fuse Two Biometrics

Biometric fusion, acquisition and combination of multiple pieces of evidence of identity, can achieve higher accuracy of biometric recognition than using a single biometric. However, fusion increases the cost or throughput of the system since it requires acquisition and processing of more samples. We document a procedure for contingent fusion. That is, two biometrics are fused only if verification on the first presented biometric is rejected. We present results of this approach for both decision and score level fusion, and examine the combination of two different algorithms, two different modalities, and two different instances of a biometric. We conclude that contingent fusion results in comparable accuracy and lower cost as measured by processing time per sample than always fusing two pieces of evidence of the same identity.