Comparative Evaluation of Biometric Identification Modalities Based on a Gaussian Score Model and Error-Rate Metrics

The paper systematises the architecture and error-rate metrics of biometric identification systems. Using a Gaussian score model, heterogeneous published FAR, FRR and EER values are converted into separability indices with confidence intervals, and the dependence of identification errors on gallery size is derived. Iris recognition shows the highest separability (d′ = 7.44), whereas estimates obtained on small samples prove unreliable. For a gallery of 106 identities, keeping FPIR at or below 1% requires FMR ≈ 10−8, which no single modality in the analysed data provides; within the Gaussian model and under the assumption of independent scores, fusion of iris and fingerprint gives a theoretical separability sufficient for this requirement (d′ ≈ 8.50). The study relies on published data, and its results require experimental confirmation on a single biometric dataset.