Research On Facial Micro-Expression Recognition Method Based On Deep Learning

Micro-expression is a dynamically changing facial expression,which is not subject to human subjective control and can reflect the most real emotions of human beings.It plays an important role in criminal investigation,medical treatment,security and other fields.However,micro-expression is characterized by short duration,weak movement intensity,and occurrence only in local areas of the face.Therefore,the research on microexpression recognition faces great challenges.With the rapid development of deep learning methods,research on micro-expression recognition has ushered in new opportunities.This thesis conducts research on micro-expression recognition based on deep learning methods,and the main research contents are as follows:1.Aiming at the complex spatiotemporal characteristics of micro-expression,the use of 3D convolutional neural networks has the problems of high computational cost and difficulty in optimization,a micro-expression recognition method based on optical flow method and pseudo-3D residual network is proposed.Firstly,an optical flow method is introduced to extract the optical flow feature sequence of micro-expression,and it is connected with the grayscale image sequence by channel to highlight the key features and provide efficient inputs for the network.Secondly,a pseudo-3D residual network based on 3D residual network is designed to model the spatiotemporal features of microexpression by using the decomposition technology of 3D convolution kernel.Meanwhile,data enhancement strategy is introduced to expand the data sample size to alleviate the over-fitting of network.Finally,the optical flow method is combined with the pseudo-3D residual network,and the extracted features are input into the classification function for emotional classification.Simulation experiments are carried out on micro-expression datasets,and the experimental results show that the proposed method can effectively extract the spatiotemporal features of micro-expression,thus improving the recognition performance of micro-expression.2.Aiming at the problem that micro-expression only occur in the local areas of the face,and their motion direction,size and degree of deformation are different,a microexpression recognition method based on optical flow characteristics and dual-attentional mechanism is proposed.Firstly,the optical flow method is introduced to extract the optical flow characteristics between the initial frame and the vertex frame of microexpression,and the corresponding optical strain is calculated to model the subtle motion information of micro-expression.Secondly,feature extraction and fusion of multi-scale informations are implemented using Inception network as the infrastructure.A dualattention mechanism is designed in Inception network to learn the key features of microexpression,and different weights are assigned to the channel information and spatial position of features through the dual-attention mechanism.So that the network adaptively learns the distinctive features of micro-expressions.Finally,the extracted features are input into the classification function for emotional classification.Simulation experiments are carried out on micro-expression datasets,and the experimental results show that the proposed method can effectively capture the key region features of micro-expression,thus improving the recognition performance of micro-expression.