Biometrics in a data stream context
Biometric systems can provide safer authentication. However, biometric features may change over time, impacting the recognition performance due to outdated biometric references. It raises the need to automatically adapt the references over time, by using adaptive biometric systems. This thesis studied several aspects of adaptive biometric systems in a data stream context. Based on this investigation, it was observed that the best choice for each aspect can be user dependent. This motivated the proposal of a modular adaptive biometric system, which can select a different configuration for each user. It also generalizes several baselines and proposals into a single modular framework, while opening numerous opportunities for future work.