We document methods for the quantitative evaluation of systems that produce a scalar summary of a biometric sample¿s quality. We predicate this on the idea that the quality measure predicts performance, whether by design or correlation. We do this abstractly (that is, for arbitrary biometrics) for the general case of a generic black box apparatus that takes a biometric sample as input, and produces some summary output that is intended to be indicative of the expected matching performance from the sample when compared with other samples. We motivate this by reviewing the valuable operational uses of quality values. We detail the quality-performance relationship, and giving various performance target measures. This is a necessary preamble to the main points of the paper: methods and metrics for evaluating a quality algorithm; and a procedure for establishing target quality values for a reference biometric data set.
Journal of Pattern Recognition and Machine Intelligence
and Tabassi, E.
Performance of Biometric Quality Measures, Journal of Pattern Recognition and Machine Intelligence, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=150605
(Accessed February 26, 2024)