Multiclass Voluntary Facial Expression Classification based on Filter Bank Common Spatial Pattern

This paper investigates the classification of voluntary facial expressions from electroencephalogram (EEG) and electromyogram (EMG) signals using the Filter Bank Common Spatial Pattern (FBCSP) algorithm. The FBCSP algorithm is an autonomous and effective machine learning approach for classifying two classes of EEG measurements in motor imagery-based Brain Computer Interface (BCI). However, the problem of facial expression recognition typically involves more than just two classes of measurements. Hence, this paper proposes an extension of FBCSP to the multiclass paradigm using a decision threshold-based classifier for classifying facial expressions from EEG and EMG measurements. A study is conducted using the proposed Multiclass FBCSP on 4 subjects who performed 6 different facial expressions. The results show that the Multiclass FBCSP is effective in classifying multiple facial expressions from the EEG and EMG measurements.