Micro-expression recognition is important because in order to recognize human’s facial expressions, changes in expressions must be immediately captured. However, this field is considered a subset of facial expression recognition (FER). The range of sequences in which human emotions, including semantic attributes, are revealed through facial expressions is extremely limited. So, it is difficult for computers to detect these subtle changes.
It seems that a full-fledged study on micro-expression recognition is being carried out, led by the Transformer family, which recently showed an amazing understanding of feature representations. The following introduces noteworthy micro-expression recognition recent studies.
[Feature Representation Learning with Adaptive Displacement Generation and Transformer Fusion for Micro-Expression Recognition]
TL;DR. They propose a method for performing feature learning from sample pairs consisting of the so-called ‘onset’ frame, where facial expression changes begin, and the ‘apex’ frame, where emotions are revealed at their maximum. An in-depth analysis of facial action units (AUs) is performed from the proposed Transformer fusion module.
[Micron-BERT: BERT-based Facial Micro-Expression Recognition]
TL;DR. They propose a method for learning attention feature representations associated with fine-grained facial expressions from frames in which significant changes are captured within sequence frames. The framework proposed by them can perform not only noise handling but also micro-movements localization.
For more detailed FER trends, please refer to the following link.
