Micro-expression Recognition Based on Residual Network
Zhang Haiwei, Shen Zhuofan · 2023
Due to the small movement amplitude of micro-expressions, the paper views micro-expression recognition as an image classification task and aims to identify micro-expressions by classifying images. Firstly, micro-expression videos are extracted into frames, each frame representing a standalone image. Secondly, each micro-expression image is fed into a neural network for feature learning and classifier training. Finally, the model is deployed for inference and recognition. The advantage of this method is that it allows the model to focus more on the location information of each micro-expression relative to the overall facial area, distinguish the shape of each AU on the human face, and identify the corresponding micro-expression category based on different AU combinations. As a result, micro-expression recognition is achieved. The test set accuracy of the model with pre-trained parameters is 3.48% higher than that of the model without pre-trained parameters, indicating that the transfer learning method used in this method can help reduce training time and improve model accuracy.