Combining biometric matchers by means of machine learning and statistical approaches

In this paper we study both machine learning and statistical approaches for combining fingerprint matchers of the FVC2006 competition. We investigate not only which is the best fusion approach, but also the correlation among the state-of-the-art matchers for fingerprint verification and scanner interoperability of the fusion techniques. Several tests are performed on all the four FVC2006 datasets, using a leave-one-out dataset testing protocol, i.e., the training phase is conducted on the datasets not used in the testing phase, so it is possible to study the pros and cons of machine learning and statistical approaches when different scanners are used in the training and testing phases.This work confirms that the fusion of different state-of-the-art fingerprint matchers can lead to a significant performance gain with respect to a single matcher.