Authors : G. Charlyn Pushpa Latha and M. Mohana Priya
Abstract: This study attempts to state that statistical signal processing treats signals as stochastic processes. It deals with the statistical properties to process signals and extract significant features. Being a versatile feature extraction method, it is also used in different areas such as natural language processing, bio-signal processing and sonar. In this research, it has been examined that the Facial Electromyography signals (FEMG) are processed by applying the statistical features in order to extract features for categorizing six emotions namely, happy, fear, neutral, sad, disgust and anger. Twenty subjects have taken part in this experimental study. The statistical features namely, kurtosis, skewness, moment, range, median absolute deviation and mean have been used to derive the significant features. Six emotions have been identified by applying the statistical features as input to neural network models. There are four neural network models namely, Cascade network, Elman network, Layered recurrent network and feed forward network have been used and compared to identify an efficient network for emotional identification. The performances of the networks in identifying the six emotions were in the range of 87.56-98.33%.
G. Charlyn Pushpa Latha and M. Mohana Priya, 2016. A Study on Emotional Identification Using Facial Electromyogram Signals and Neural Networks. International Journal of Soft Computing, 11: 437-443.