Micro-Expression Recognition Based on MAML Meta-Learning Algorithm

Micro-expression, an accurate and effective behavioral clue, can reveal true changes in people’s emotions. Nowadays, micro-expression recognition is widely used in education, interrogation, clinical diagnosis, etc. However, previous research on micro-expression recognition focused on deep-learning methods that depend on large amounts of training samples. Recently, meta-learning mainly improves performance by extracting meta-knowledge from a set of tasks and has been widely used in few-shot classification fields. In this paper, we propose a micro-expression recognition method based on Model-Agnostic meta-learning algorithm. This method meta-learned the initialization parameters of a three-dimensional convolutional neural network (3DCNN) to achieve fast adaptation to new tasks. We conducted experiments on two typical micro-expression datasets: CASME II and SMIC-HS. The results showed that the recognition accuracy of our proposed method can reach 92.97%, which significantly outperforms the state-of-the-art algorithms. Furthermore, the recognition accuracy of our approach is independent of the number of training samples.