Facial Expression Recognition Based on Semi-supervised Vision Transformer
Xiangwei Fu · 2022
The purpose of Facial expression recognition is to classify human expression using the designed algorithm. FER2013 database which contains 30,000 facial expressions in 7 groups is an emerging human emotion database used in training models. Furthermore, semi-supervised learning is a data processing technique, which is gradually gaining importance in natural language processing with the support of computer technology. Semi-supervised combines supervised learning and unsupervised learning. It learns from a small number of labeled examples to predict a large amount of unlabeled data, moreover, the model must learn and predict new examples. The basic process of semi-supervised learning involves using the labeled data to mark the unlabeled data, to significantly help to increase the training data. This paper aims to design a Vision transformer (ViT) semi-supervised learning model to classify facial expressions based on FER2013 and achieved an 89.43% high accuracy.