Micro-Expression Recognition Base on Optical Flow Features and Improved MobileNetV2

When a person tries to conceal emotions, real emotions will manifest themselves in the form of micro-expressions. Research on facial micro-expression recognition is still extremely challenging field pattern recognition. This because it difficult implement best feature extraction method cope with micro-expressions small changes and short duration. Most methods are based hand-crafted features extract subtle movements. In this study, we introduce that incorporates optical flow deep learning. First, take out onset frame apex from each video sequence. Then, motion between these two frames extracted using method. Finally, inputted into an improved MobileNetV2 model, where SVM applied classify expressions. order evaluate effectiveness method, conduct experiments public spontaneous database CASME II. Under condition applying leave-one-subject-out cross-validation accuracy rate reaches 53.01%, F-score 0.5231. The results show proposed can significantly improve performance.