Micro-expression Recognition Based on Apex Frame Selection Strategy and Optical Flow Features

Micro-expressions are generated by real emotions and are difficult to fake. The movements of micro-expressions are localized and the intensity of movements is weak. It is hard for current algorithms to accurately and quickly recognize micro-expressions in video clips. Therefore, we propose a micro-expression recognition method based on an apex frame selection strategy and optical flow features to address this issue. The method utilizes an apex frame extraction strategy based on changes in image optical flow information to select the start frame and the apex frame that can characterize the entire video. Simultaneously, the optical flow information features that characterize the start frame and the apex frame are preserved. The optical flow features extracted are then used as input information for the ResNet18 network for classification. The approach we propose is able to better capture the spatial features of facial micro-expressions and improve the accuracy of micro-expression recognition. The method suggested in this paper holds significant potential for various applications in fields such as psychotherapy and disease recognition.