Pengenalan Ekspresi Wajah dengan ResNeXt

Abstract

Facial expressions are movements or positions in human facial muscles that are one way humans convey their emotional states. Facial expression recognition is one of the important fields in computer vision because automatic and accurate facial expression recognition will greatly help in improving the quality of interaction between humans and computers, especially in the fields of psychology, criminal investigation, and security. In this study, the author aims to develop a facial expression recognition model using the Convolutional Neural Network method and the ResNeXt architecture that is able to produce deeper feature representations through the cardinality mechanism. The dataset used in this study is the FER+ dataset that provides annotations with a crowdsourcing mechanism. The dataset consists of eight classes of facial emotions, namely neutral, happiness, surprise, sadness, anger, disgust, fear, and contempt. The results of the model test show an accuracy of 82% with a macro average f1-score of 68% and a weighted f1-score of 82%.