LAUSR

Abstract The exploitation of brain signals for biometric recognition purposes has received significant attention from the scientific community in the last decade, with most of the efforts so far devoted to the quest for discriminative information within electroencephalography (EEG) recordings. Yet, currently-achievable recognition rates are still not comparable with those granted by more-commonly-used biometric characteristics, posing an issue for the practical deployment of EEG-based recognition in real-life applications. Within this regard, the present study investigates the effectiveness of deep learning techniques in extracting distinctive features from EEG signals. Both convolutional and recurrent neural networks, as well as their combinations, are employed as strategies to derive personal identifiers from the collected EEG data. In order to assess the robustness of the considered techniques, an extensive set of experimental tests is conducted under very challenging conditions, trying to determine whether it is feasible to identify subjects through their brain signals regardless the performed mental task, and comparing acquisitions collected at a temporal distance greater than one year. The obtained results suggest that the proposed networks are actually able to exploit the dynamic temporal behavior of EEG signals to achieve high-level accuracy for brain-based biometric recognition.