Assessment of Synthetically Generated Mated Samples from Single Fingerprint Samples Instances
Simon Kirchgasser, Christof Kauba, Andreas Uhl · 2021
The availability of biometric data (here fingerprint samples) is a crucial requirement in all areas of biometrics. Due to recent changes in cross-border regulations (GDPR) sharing and accessing biometric sample data has become more difficult. An alternative way to facilitate a sufficient amount of test data is to synthetically generate biometric samples, which has its limitations. One of them is the generated data being not realistic enough and a more common one is that most free solutions are not able to generate mated samples, especially for fingerprints. In this work we propose a multi-level methodology to assess synthetically generated fingerprint data in terms of their similarity to real fingerprint samples. Furthermore, we present a generic approach to extend an existing synthetic fingerprint generator to be able to produce mated samples on the basis of single instances of non-mated ones which is then evaluated using the aforementioned multi-level methodology.