Preprocessing data for facial gestures classifier on the basis of the neural network analysis of biopotentials muscle signals

Abstract

The recognition of facial gestures using biopotentials muscle signals has been proposed for human machine interface. Real-time myoelectric control requires a high level of accuracy and computational load, so a compromise between these two main factors should be considered. The most informative electromyogram features, required number of channels and the most suitable architecture of the neural network were identified in this study. In this paper, a results of preprocessing data were proposed to use in facial gestures classifier. The effectiveness of different sets input data combinations was also explored to introduce the most discriminating.