Multi-order biometric score analysis framework and its application to designing and evaluating biometric systems for access and border control
Dmitry O. Gorodnichy · 2011
Traditionally, automated access and border control biometric systems are thought of and designed as verification 1-to-1 systems, where a single comparison between a probe and the claimed identity is examined to allow or disallow the entry to a person; and as such they have been evaluated to date - by using the error tradeoff statistics, which counts how many times a person was falsely accepted or rejected. Such a design however may soon become obsolete due to the recent shift towards applying biometrics to free-flow surveillance-like environments and also in the light of recent findings showing that performance of many verification systems can be improved through the use of several 1-to-N scores, instead of relying on a single 1-to-1 score only. As the framework for designing biometric-enabled access and border control systems changes, so has to change the methodology for the evaluation of such systems. This paper addresses this problem by establishing the multi-order biometric score analysis framework. The framework incorporates latest innovations and recommendations related to the comprehensive evaluation of biometric systems, including subject-based analysis, calibrated score analysis, and two new performance metrics: threshold-validated recognition ranking and non-confident decisions due to multiple threshold-validated scores. The framework is implemented in the Comprehensive Biometrics Evaluation Toolkit (C-BET) and has been applied for the evaluation of several biometric modalities, in particular, those that are frequently contemplated for the use in unconstrained access-border control applications, such as face, voice and iris. The results of the iris modality evaluation are presented in this paper.