Comparison of different time-domain feature extraction methods on facial gestures' EMGS

Electromyography is a bio-signal which is applied in various fields of study such as motor control, neuromuscular physiology, movement disorders, postural control, human ma-chine/robot interaction and so on. Processing of these bio-signals is the essential fact during each application and there stil...

Full description

Saved in:
Bibliographic Details
Main Authors: Hamedi, Mahyar, Salleh, S. H., Mohd. Noor, Alias, Afizam, I. K.
Format: Article
Language:English
Published: 2012
Subjects:
Online Access:http://eprints.utm.my/id/eprint/46712/1/M.Hamedi_2012_Comparison%20of%20different%20time-domain%20feature%20extraction%20methods%20on%20facial%20gestures%27%20EMGS.pdf
http://eprints.utm.my/id/eprint/46712/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Electromyography is a bio-signal which is applied in various fields of study such as motor control, neuromuscular physiology, movement disorders, postural control, human ma-chine/robot interaction and so on. Processing of these bio-signals is the essential fact during each application and there still can be seen many challenges among researchers in this area. This paper is focused on the comparison between the classification performances by using different well known feature extraction methods on facial EMGs. Totally ten facial gestures namely smil-ing with both side of lips, smiling with left side of lips, smiling with right side of lips, opening the mouth like saying 'a' in apple word, clenching the molar teeth, gesturing 'notch' by raising the eyebrows, frowning, closing the both eyes, closing the right eye and closing the left eye are recorded from 6 participants through 3 bi-polar recording channels. In the first step, the signals are filtered to get prepared for better processing. Then, time-domain feature extraction methods INT, MAV, MAVS, RMS, VAR, and WL are applied to signals. Finally, the features are classified by Fuzzy C-Means in order to achieve the recognition accuracy and evaluate the performance of each feature extraction method. This work is carried out by revealing that, RMS gives the most probability amplitude approximation in a steady power and non-tiring contraction when the sig-nal is modeled as Gaussian random process. In contrary, WL proved its weakness in estimating the value of facial EMGs.