A real-time finger vein verification system with improved erollment design

Abstract

In practical applications, enrollment is an indispensable part of a biometric system. Most of the existing literature in finger vein recognition primarily focuses on the design of feature extraction methods, and the investigation of enrollment strategies is rarely seen. A proper enrollment scheme can improve the enrolled finger vein identity's representation, thereby boosting the final recognition performance. In this work, we perform a systematic investigation of different enrollment strategies for finger vein-based biometric recognition. We studied two types of enrollment schemes, single-template scheme and multi-template scheme. Experimental results showed that the multi-template scheme has significantly improved the equal error rate in simulated real-life applications compared to the single-template scheme. We also investigated the cost of time, comparing single-template schemes and multi-template schemes. To conduct the experiments more effectively, a Graphical User Interface (GUI) program has been developed to collect finger vein video, extract finger vein feature, enroll the feature to the database, and real-time finger vein verification.