Using Chebyshev's inequality to determine sample size in biometric evaluation of fingerprint data
Jin Chu Wu, Charles L. Wilson · 2005
The fingerprint datasets in some cases may exceed a million of samples.The underlying distribution functions of the similarity scores are unknown.Therefore, the needed size of a biometric evaluation test set is an important question in terms of both efficiency and accuracy.In this article, Chebyshev's inequality, in combination with simple random sampling, is used to determine the sample size for biometric applications.The performance of fingerprint-image matcher is measured by both the area under a Receiver Operating Characteristic (ROC) curve and the True Accept Rate (TAR) at an operational False Accept Rate (FAR).The Chebyshev's greater-than-95% intervals of these two criteria based on 500 Monte Carlo iterations are computed for different sample sizes as well as for both high-and low-quality fingerprint-image matchers.The stability of such Monte Carlo calculations with respect to the number of iterations is also presented.The choice of sample size is dependent on the qualities of fingerprint-image matchers as well as on which performance criterion is invoked.However, in general, for 6000 match similarity scores, 50000 to 70000 scores randomly selected from 35994000 non-match similarity scores can ensure reasonable accuracy with greater-than-95% probability.