Searching form the Model of Human Facial Expressions and its Implementation inMATLAB and Simulink Environments

Abstract

The publication contains selected research results on the creation of a neural model of human facial expressions along with its implementation in the MATLAB and Simulink environments using the library called Deep Learning Toolbox. The model was generated as an object to recognize selected human faces and emotional states based on visual data recorded on the human face in real time. The study was placed against the background of available literature on the analysis of facial expressions and emotion classification methods. In addition to the concept of the original solution, the assumptions of the research experiment were given, a method for measuring facial expressions was selected for the experimental conditions, and a set of data was developed for training a neural model of the facial expression system with their preparation for ANN learning. Ultimately, various artificial neural networks were trained to model the facial expression system and sensitivity and comparative tests were performed in Simulink to, among others: assess the quality of the model in relation to real data. Very high results of ANN training of the facial expression system model were obtained (MSE on the order of 10−14, R close to 1), as well as relatively high model quality relative the facial expression system, measured, among others, average relative error, the value of which was several percent. However, in terms of prediction effectiveness, because of the obtained results, which are not very high for the assumed high measurement accuracies, the research that has been started is continued.