A Review of Deep Learning-Based Micro-Expression Classification

Abstract

In daily life, it is important to discern the emotions of others. People always rely on facial expressions and body language to discern the emotional tendencies of others, so that they can choose appropriate ways to communicate and interact with others. Macro expression has a large range of action, while micro expression has a small range of action and is not easy to detect. The former can be falsified while the latter are generated spontaneously. Therefore, analyzing micro-expressions can more accurately determine a person's true emotions, which can then assist criminal investigation or provide psychological therapy. However, even trained individuals find it challenging to correctly identify micro-expressions because of their fleeting and nuanced characteristics, hence, machine learning techniques are frequently employed to handle such tasks. This article looks into deep learning and conventional machine learning algorithms for micro-expression classification. Sentiment analysis methods based on machine learning usually detect facial changes based on texture features and optical flow features. Deep learning methods can be used to complete end-to-end classification through modules such as CNN, GCN and transfer learning. Finally, it was found that to further improve classification accuracy, enhancing the quality of datasets and attempting to apply self-supervised methods to train deep models could be beneficial.