Single trial authentication with mental password writing

Abstract

This paper presents an authentication system that uses brain waves as a biometric discriminant trait. It utilizes Electroencephalogram (EEG) signals generated from mental writing of the user-owned password. Independent Component Analysis (ICA) and baseline correction has been used for preprocessing and noise removal. The effect of two types of features, multivariate autoregressive (MVAR) model parameters and power spectral density (PSD) features, have been studied for this activity. Performance results based on single trial analysis have revealed that imagined password writing can reach average Half Total Error Rate (HTER) of 5% for PSD features vs 3% obtained with MVAR coefficients. The experiments have shown that mental password writing can be used for increasing the user acceptance for enrollment conditions while maintaining high performance results.