Multi-Biometric Identification With Cascading Database Filtering

The growing scale and number of biometric deployments around the world necessitates research into technologies which facilitate fast identification queries and high discriminative power. In this context, this article presents a biometric identification system which relies on a successive pre-filtering of the potential candidate list using multiple biometric modalities, coupled with a weighted score-level information fusion. The proposed system is evaluated in a series of experiments using a compound dataset constructed from several publicly available datasets; an open-set identification scenario is considered with the enrolment database containing 1,000 chimeric instances. This evaluation shows that the proposed system exhibits a significantly increased biometric performance w.r.t. a weighted score-level or rank-level fusion based baseline, while simultaneously providing a consequential computational workload reduction in terms of penetration rate. Lastly, it is worth noting that the proposed system could be flexibly employed in any multi-biometric identification system, irrespective of the chosen types of biometric characteristics and the encoding of their extracted features.