Exploiting the “doddington zoo” effect in biometric fusion

Abstract

Recent research in biometrics has suggested the existence of the “Biometric Menagerie” in which weak users contribute disproportionately to the error rate (FAR and FRR) of a biometric system. The aim of this work is to utilize this observation to design a multibiometric system where information is consolidated on a user-specific basis. To facilitate this, the users in a database are characterized into multiple categories and only users belonging to weak categories are required to provide additional biometric information. The contribution of this work lies in (a) the design of a selective fusion scheme where fusion is invoked only for a subset of users, and (b) evaluating the performance of such a scheme on two public datasets. Experiments on the multi-unit CASIA V3 iris database and multi-unit WVU fingerprint database indicate that selective fusion, as defined in this work, improves overall matching accuracy while potentially reducing overall computational time. This has positive implications in a large-scale system where the throughput can be substantially increased without compromising the verification accuracy of the system.