Fast Micro-Expression recognition method based on Bi-Directional optical flow

Micro-expressions contain rich emotional and cognitive information, and can serve as important indicators of mental status. Systematic observation and precise recognition of facial micro-expressions in the elderly can provide critical and valuable clues for early diagnosis of dementia, dynamic monitoring of the disease, and evaluation of treatment effects. However, the micro-expression motion intensity is weak, and the traditional optical flow-based micro-expression recognition method is prone to extracting erroneous optical flow information, which affects the recognition accuracy of the system. This paper proposes a Bi-directional Optical flow-based Fast Micro-expression recognition method (BOFM), which employs forward and reverse bi-directional optical flow to capture facial movements and accurately recognize micro-expressions. Additionally, it introduces a keyframe extraction method that utilizes the variation effect of the optical flow field to eliminate unnecessary frames in video clips and enhance the real-time performance of the system. This method has been validated using public datasets such as SMIC and CASME. The verification results demonstrate that this method achieves approximately a 9.3% higher accuracy compared to the Sparse MDMO (Main Directional Mean Optical-flow) algorithm. Notably, the proposed algorithm showcases a significant running time reduction of approximately 36.5% when compared to the micro-expression recognition algorithm based on FlowNet2. These findings clearly indicate that the proposed algorithm possesses excellent capabilities in recognizing micro-expressions. This establishes a basis for future investigations into the connection between facial micro-expression patterns and early-stage Alzheimer's disease, playing a crucial role in its early detection and prevention.