Exploring P300-Based Biometric For Individual Identification Based On Convolutional Neural Networks

The potential of using the electrical brainwave signals of individual’s neural response to stimuli (the event-related potential) as a biometric in subject identification has been investigated. Electroencephalography (EEG) signals from 24 participants actively involving in the P300 Speller task are used to develop biometric systems based on discriminative classifiers. P300 is an event-related potential (ERP) component in human EEG elicited using the oddball stimulus to reflect the individual’s reaction in a target detection process [1]. For P300, it is possible to extract unique neural response pattern and information from different subjects to determine the subjects’ identity. Biometric recognition based on neural response pattern could be a physiological characteristic. Thus, while P300 inherit the advantages of human physiological features as a mean of individual identification, it is hard to steal, or replicate compared to other physiological features (e.g. fingerprint, iris). This abstract explores the possibility of using P300-based biometric as an individual identification tool. Eight-channel EEG data were recorded, and band-pass filters were applied to remove artifacts and to reduce noise. Topographic plot was used for feature extraction and convolutional neural net (CNN) was applied for classification. SVM and ELM were also used as classifiers.