Micro-expression Detection based on Action Units and Multi-region Feature Fusion
Hao Sun, Bing Li, Lei Chen, Jun Yao, Xianye Ben · 2022
Recently, micro-expression detection has attracted much attention, since micro-expression has potential lie detection applications. In this paper, a micro-expression detection algorithm based on action units (AUs) and multi-region feature fusion is proposed. First, the regions of interest (ROIs) are selected based on AUs. Second, the optical flow features in spatio-temporal domain and frequency domain features are fused as a final feature descriptions of micro-expressions. Finally, a decision criterion based on the characteristic of the optical flow angle, magnitude and frequency magnitude is applied to spot onset, apex, and offset frames from micro-expression video sequences. The experiments are performed on SDU_spotting and CASME II datasets. The results show that the proposed method obtains higher performance than the state-of-the-art methods in terms of accuracy.