Brainprint Recognition Based on the Stable SSVEP Space-Frequency Energy Distribution

Brainprint recognition has received increasing attention in information security. Electroencephalography (EEG) signals measured under task-related or task-free conditions have been exploited as brain biometrics. However, what components make the uniqueness of one's brain signals remains unclear. In this study, we proposed an interpretable biomarker based on steady-state visual evoked potentials (SSVEP) signals for EEG biometric identification. Firstly, we recovered pure SSVEP components from EEG by a point-position equivalent reconstruction (PPER) method. Then, we calculated the distribution properties of SSVEP components in space and frequency. By using the uniform manifold approximation and projection, we reduced the distribution features to 2-dimensions, which shows the separability of the subjects. Lastly, we built a long short-term memory (LSTM) network to perform brainprint recognition on the SSVEP benchmark dataset. The average recognition accuracy can reach up to 98.33%. Our results demonstrate that the space-frequency energy feature of SSVEP is an effective and interpretable biomarker for brainprint recognition. This study provides a further understanding of the uniqueness of individual EEG signal, and facilitates its potential application for personal identification.