Evolution and evaluation of biometric systems

Biometric systems have evolved significantly over the past years: from single-sample fully-controlled verification matchers to a wide range of multi-sample multi-modal fully-automated person recognition systems working in a diverse range of unconstrained environments and behaviors. The methodology for biometric system evaluation however has remained practically unchanged, still being largely limited to reporting false match and non-match rates only and the tradeoff curves based thereon. Such methodology may no longer be sufficient and appropriate for investigating the performance of state-of-the-art systems. This paper addresses this gap by establishing taxonomy of biometric systems and proposing a baseline methodology that can be applied to the majority of contemporary biometric systems to obtain an all-inclusive description of their performance. In doing that, a novel concept of multi-order performance analysis is introduced and the results obtained from a large-scale iris biometric system examination are presented.