Analysis of familiar and unfamiliar images using power spectral estimation for EEG authentication system

Biometric authentication is a system used for recognizing an individual according to their physiological and behavioral characteristics, Recently, the application of biometric authentication is the most useful in cybersecurity such as fingerprints, facial, and voices. Traditional authentication such as password and PIN have been used for a decade, however, they bring drawbacks to the users which were attacked by cybercriminals. Therefore, the brainwave of electroencephalogram (EEG) is proposed as the biometric trait to encounter the problems faced. The aim of this study is to explore the feature extraction method by applying the power spectral estimation method as linear feature analysis such as the Welch method, Burg Method, Yule-Walker method, Covariance method and Modified Covariance method. After extracting five features, the statistics of mean, median, variance, and the standard deviation is computed and fed into three types of shallow classifiers including the Neural Network, K-Nearest Neighbors, and Support Vector Machine. As a result, the Yule-Walker feature contributes to the highest average accuracy and the Neural Network using Levenberg-Marquardt (LM) has achieved the highest accuracy for all frequency bands. In fact, the highest frequency band obtained by gamma (30-45 Hz) followed by highbeta (20-30 Hz), lowbeta (13-20 Hz) and alpha (8-13 Hz). Overall, all three features and three classifiers are able to achieve between 70.1% to 98.2% of accuracy which shows that it can differentiate between different tasks.