Relationship between Number of Subjects and Biometric Authentication Equal Error Rates
share
The relationship between the number of subjects included in a biometric authentication study and biometric performance assesed with equal error rate (EER) is not well studied. In the present paper, we use both synthetic and real data to study this relationship. Although we and others had hypothesized that more information would be required to authenticate more subjects at a given EER, our report does not support this hypothesis, with rare exception. With synthetic data sets of N = 1,000, N= 10,000 and N = 100,000 we found remarkably similar (and specifically not statistically different) EERs. In a real face recognition data set, with data subsets of N = 1,000 subjects and N = 10,000 subjects, we also found very similar EERs. Our evidence indicates that EER does not change as a function of number of subjects, at least in the range of subject numbers studied herein.
READ FULL TEXT