Improved EEG based human authentication system on large dataset

The genetic traits in the Electroencephalogram (EEG), has made it a possible characteristic for the continuous biometric human verification. It is important for a biometric characteristic to have an acceptable performance on large populations besides satisfying the main requirements of a biometric system such as uniqueness, universality, acceptability, permanence, collectability, circumvention and measurability. In the present work we have proposed an EEG based human authentication system on a large dataset containing 104 healthy subjects; an issue which has not been focused in the literature. The method has been examined in two situations, closed and opened eyes. By extracting Autoregressive (AR) coefficients as the feature set, selecting the features using a statistical based method, doing a two-step feature normalization and selecting reference sweeps, the proposed system achieved to 97.43±0.00% accuracy using only 10% of the data in the training step.