Research on Improved CNN and Dual-Layer LSTM Micro-expression Recognition

Micro-expressions are brief and subtle facial expression changes in humans, which typically occur in a very short time and reflect people’s unconscious true emotions. The transient and delicate characteristics of this type of expression pose great challenges for automatic detection and recognition. In recent years, how to analyze human facial micro-expressions through artificial intelligence has become one of the research hotspots. This paper aims to design and implement a micro-expression recognition system based on deep learning, by combining convolutional neural networks (CNN) and long short-term memory networks (LSTM) to effectively capture the spatial features and temporal dynamics of micro-expressions. An improved CNN model is adopted to extract the spatial features of facial images, and the LSTM network is utilized to capture the subtle changes of these expressions over time, in order to recognize the briefly appearing micro-expressions. The experimental results on the CASMEII dataset show that the proposed CNN and dual-layer LSTM model significantly outperforms traditional methods in micro-expression recognition, with an accuracy of 78.90%.