Research And Implementation Of Micro-expression Recognition Algorithm Based On Graph Convolutional Neural Network

Micro-expression is a short-lived facial expression that humans make unconsciously when trying to hide an emotion and it lasts only 0.04 to 0.2 seconds.In contrast,expression with a duration of 0.75 seconds to 2 seconds are called macro-expression.Micro-expression is a spontaneous facial expression that can not be controlled by humans.Therefore,microexpression can more accurately express people’s true emotion than macro-expression.Based on the above characteristics,micro-expression has important reference value in the diagnosis of mental illness and the interrogation of major criminal cases.Traditional micro-expression recognition methods first extract hand-crafted features and then use machine learning algorithms for classification.These methods rely heavily on hand-crafted features and generally have low classification accuracy.Deep learning methods based on convolutional neural networks have been widely used in microexpression recognition.Compared with traditional methods,the accuracy of microexpression recognition has been greatly improved.However,limited by the short duration and low intensity of micro-expression actions,the deep learning micro-expression recognition algorithm still needs to be further improved.Based on the above characteristics of micro-expression,this dissertation conducts research in the following three aspects:(1)A single-input,single-output network model is proposed.The micro-expression optical flow sequence is input.The facial action unit features are initially obtained through a 3D convolutional neural network and the facial action features are further obtained through a graph convolutional network based on the correlation of facial action units.Finally,the micro-expression classification is performed according to the acquired facial action unit features.(2)A dual-input,single-output network model is proposed.Micro-expression video sequences and micro-expression optical flow sequences are the input.The facial action unit features are obtained from the micro-expression optical flow sequence as the facial local information and the overall facial information is obtained from the micro-expression video sequence.Finally,the overall information and local information of the face are fused to perform micro-expression classification.(3)By applying the micro-expression recognition algorithm based on graph convolutional neural network proposed in this dissertation,a micro-expression recognition system is designed.The two micro-expression recognition network models proposed in this dissertation have been tested on two public datasets SMIC and CAMSE II and achieved good results.