Human identification with EEG signals in different emotional states

This paper investigates human identification using EEG signals. It has been shown that Electroencephalogram (EEG) can be used as a trait for biometric systems. Previous studies have reported proper channels and features in resting states and mental tasks. However, since EEG signal is sensitive to emotion, the stability of reported features during emotional states is not well verified. Our goal is to investigate channels and features which have stable results regardless of emotional states. To this end, three experiments were designed: `training' and `testing' an identification system with 1) mixture of emotional states; 2) the same specific emotional states; 3) different emotional states. 1728 features were extracted which later construct the feature vector of each subject and then Support Vector Machine (SVM) was used to classify the subjects. After selecting 5 best features, the Correct Classification Rate (CCR) is in the range of 88% to 99% for 3 experiments. Moreover, we found that features extracted from Gamma frequency band in Left-Posterior quarter of the brain have more stable and reliable information for human identification, regardless of emotional states, comparing to other features.