Classification of EEG signals of familiar and unfamiliar face stimuli exploiting most discriminative channels

Abstract

The objective of the study is to classify electroencephalogram signals recorded in a familiar and unfamiliar face recognition experiment. Frontal views of familiar and unfamiliar face images were shown to 10 volunteers in different sessions. In contrast to previous studies, no marker button was used during the experiment. Participants had to decide whether the displayed face was familiar or unfamiliar at the instant of stimulus presentation. The signals were analyzed in the preprocessing, channel selection, feature extraction, and classification stages. The novel two-feature extraction and eight-channel selection methods were applied to the analyses. Sixteen classification results were compared and the best performance was investigated. Consequently, the highest average classification accuracy was obtained at 72.67% when piecewise constant modeling feature extraction and relative entropy channel selection methods were used.