Grammatical Facial Expression Recognition Based on Machine ‎Learning

Abstract

 Facial expression recognition is an evolving field of research with various applications in system development. Recognizing facial expressions is particularly crucial in discourse construction. This study focuses on the design of a grammatical facial expressions (GFEs) recognition system that depended on extracted features from the estimation of head pose and detection of Action Units (AUs) in order to recognize grammatical expressions. The Facial Action Coding System (FACS) is utilized to depict AUs, which effectively capture and classify the intricate movements of facial muscles. Among the AUs, AUs 1 and 4 serve as potential indicators for recognizing grammatical expressions. The process of estimating head pose produces characteristics, including Euler angles (namely, pitch, roll, and yaw), as well as 3D coordinates, which signify the relative arrangements of facial landmarks in correspondence to the camera. The present study employs a dataset comprising video recordings obtained from a sample of 53 individuals whose ages range from 18 to 44 years. Two distinct classifications, namely Multilayer Perceptron (MLP) and K-Nearest Neighbor (KNN), are employed in the final stage of the proposed system. The experimental outcomes show that the KNN classifier attains better efficacy in contrast to the MLP classifier.