Multifinger Penetration Rate and ROC Variability for Automatic Fingerprint Identification Systems

AFIS system design and performance is heavily impacted by data correlations and error rate variability both within and across individual fingerprint records, but little is known about these issues. In this paper, we report on experiments done with the best algorithms from six major AFIS vendors tested using a 4128 × 4080 database of electronically collected flat prints. We obtain Receiver Operating Characteristic curves for thumb through ring finger on right and left hands, as well as experimental penetration and binning error rates for one-, two-, four-, and eight-print systems. Impact of binning on the overall ROC curve is measured. “Zero-effort” impostor error rate variability (“lambs” and “wolves”) is observed across the data.