Classification of Human Emotional States using Facial Electromyogram Signals

Hema C.R and G. Charlyn Pushpa Latha

Abstract:

Statistical signal processing is a method applied in signal processing for analysing and extracting information from signals and noise based on their stochastic properties. Statistical feature is an efficient feature extraction method used in diverse applications such as biosignal processing, image processing, seismic data processing, radar and sonar. In this paper, we have investigated the application of statistical features to derive a set of features from facial electromyography signals (FEMG) for classifying six emotional states namely anger, disgust, fear, happy, neutral and sad. This paper focuses on how emotional experiences are expressed in six parameters namely mean, median absolute deviation, range, moment, skewness, and kurtosis. FEMG signals were collected from 20 subjects in a controlled environment using audio-visual stimuli. Two Neural Network models namely Feed Forward Neural Network and Elman Neural Network were used to identify these six emotional states. From the results it is inferred that the mean classification accuracies of 90% and 98.33% were obtained in classifying these six emotional states.

Keywords: Electromyography, Facial Electromyography, Statistical Features, Feed Forward Neural Network, Elman Neural Network

Conference Name: International Engineering Post Graduate Research Conference

Conference Date: 12, March 2015 - 13, March 2015

Pages: 186-192

Paper ID: chapter-36