Eye blink artifact removal in eeg using tensor decomposition

Abstract

EEG data are usually contaminated with signals related to subject's activities, the so called artifacts, which degrade the information contained in recordings. The removal of this additional information is essential to the improvement of EEG signals' interpretation. The proposed method is based on the analysis, using Tucker decomposition, of a tensor constructed using continuous wavelet transform. Our contribution is an automatic method which processes simultaneously spatial, temporal and frequency information contained in EEG recordings in order to remove eye blink related information. The proposed method is compared with a matrix based removal method and shows promising results regarding reconstruction error and retaining the texture of the artifact free signal.