Structure in errors: a case study in fingerprint verification

Abstract

Measuring the accuracy of biometrics systems is important. Accuracy estimates depend very much on the quality of the test data that are used: including poor quality data will degrade the accuracy estimates. Factors that determine the good quality data and poor quality data can not be revealed by simple accuracy estimates. We propose a novel methodology to analyze how the overall accuracy estimate of a system relates to the specific quality of biometrics samples. Using a large collection of fingerprint samples, we present an analysis of system accuracy, which suggests that a significant part of the error is due to few fingers.