Research On Key Techniques Of Facial Micro-expression Recognition Based On Video

Micro-expression is a very short-lived,involuntary facial expression that is revealed when humans try to suppress or hide real emotions.It has a short duration and low intensity,and is a natural expression that is difficult to imitate.Micro-expression auto-recognition technology can be widely used in criminal investigation judicial science,clinical medicine and other fields.The development of a system for automatic recognition of micro-expressions,which enables computers to automatically detect and recognize micro-expressions,is of great significance to the research and application of micro-expression mechanisms.Video-based micro-expression recognition technology is mainly divided into two parts: key frame detection technology and micro-expression recognition technology.In the detection of the micro-expression key frame,face detection is first performed on the video,and the face area is divided into 6×6 small blocks according to the coordinates of the landmark points.The features of the 36 small blocks are extracted for feature difference analysis and peak detection is used to judge the frame with the largest motion amplitude.In the feature extraction stage,the impact of LBP,HOOF and HOG on the detection results was analyzed.The experimental results show that these three features can achieve good detection results.For micro-expression recognition technology,micro-expression recognition based on LBP-TOP features and deep learning are studied.In the LBP-TOP feature-based micro-expression recognition method,Eulerian video magnification is used to enlarge small changes in the video after face alignment,and then the face is divided into multiple regions,and LBP-TOP features are extracted in each region.Linear SVM classifier is used for classification.The experiments show that when using the LBP-TOP feature,the highest accuracy rate is 66.7% when the face block is 6×6 and the EVM magnification factor is 10.The deep-learning-based micro-expression recognition scheme uses the migration learning method,uses the fine-tuned CNN network to extract the spatial features of the micro-expression,and then the LSTM is used to extract the time-series features of the micro-expression.The program based on deep learning achieved 62.75% accuracy in the experiment.The CASME ? data set was used in the experiments,a complete micro-expression recognition system was basically implemented,and various modules in the program were analyzed,and finally a good recognition result was achieved.Although many advances have been made in the detection and recognition of micro-expressions,the distance from the actual application is still insufficient.With the expansion of future data sets and the advancement of related technologies,the micro-expression recognition technology will surely be further developed.