基于持续时空注意力网络的人脸微表情识别.

Facial micro-expressions have the characteristics of short duration, small range of motion and only occurs in local areas of the face, which brings great challenges to the accurate recognition of micro-expressions.Aiming at solving these problems, a micro-expression recognition algorithm based on continuous spatiotemporal attention network is proposed.The algorithm consists of two channels: primary and secondary.The primary channel is the continuous spatiotemporal attention module and the secondary channel is the position calibration module.Firstly, the main channel is discretely sampled, and the original video frames are extracted at equal intervals to form a new video sequence.The motion difference between each frame is extracted by the inter frame difference method, and then it is input into the continuous spatiotemporal network to extract the motion spatiotemporal features of facial muscles.Secondly, the position of the main channel information is calibrated by using the facial position information extracted by the sub channel.Finally, the classifier Softmax is input to classify the micro-expression.Experiments show that the average recognition accuracy of the algorithm is 89.96%,86.73% and 89.76% respectively on three micro-expression data sets, which is higher than other existing algorithms. [ABSTRACT FROM AUTHOR]