Quantative Evaluation of the Efficiency of Facial Bio-potential Signals Based on Forehead Three-Channel Electrode Placement For Facial Gesture Recognition Applicable in a Human-Machine Interface

Today, facial bio-potential signals are employed in many human-machine interface applications for enhancing and empowering the rehabilitation process. The main point to achieve that goal is to record appropriate bioelectric signals from the human face by placing and configuring electrodes over it in the right way. In this paper, heuristic geometrical position and configuration of the electrodes has been proposed for improving the quality of the acquired signals and consequently enhancing the performance of the facial gesture classifier. Investigation and evaluation of the electrode's proper geometrical position and configuration can be performed using two clinical and modeling. In the clinical method, the electrodes are placed in predefined positions and the elicited signals from them are then processed. The performance of the method is evaluated based on the results obtained. On the other hand, in the modeling approach, the quality of the recorded signals and their information content are evaluated only by modeling and simulation. In this paper, both methods have been utilized together. First, suitable electrode positions and configuration were proposed and evaluated by modeling and simulation. Then, the experiment was performed with a predefined protocol on 7 healthy subjects to validate the simulation results. Here, the recorded signals were passed through parallel butterworth filter banks to obtain facial EMG, EOG and EEG signals and the RMS features of each 256 msec time slot were extracted. By using the power of Subtractive Fuzzy C-Mean (SFCM), 8 different facial gestures (including smiling, frowning, pulling up left and right lip corners, left/right/up and down movements of the eyes) were discriminated. According to the three-channel electrode configuration derived from modeling of the dipoles effects on the surface electrodes and by employing the SFCM classifier, an average 94.5% discrimination ratio was obtained. The results can validate our hypothesis and the simulation results too.Discussion and Based on the obtained results, it is clear that our proposed electrode configuration and placement is an efficient method which can be used for further applications such as designing and implementing a robust human-machine interface.