Research On Facial Micro-expression Recognition Methods Based On Deep Learning

With the widespread application of AI and big data technologies,facial recognition and facial expressions recognition have received more and more attention.Facial expressions are an important way for humans to convey emotions.Recognition of facial expressions has always received great attention and has been widely used in many fields.At present,the recognition effect of facial expressions meets daily and social needs to a certain extent.However,due to factors such as low resolution,light interference,image occlusion,and weak facial expressions,facial expressions are not easy to detect in addition to macro expressions that have obvious external expressions in daily life.The slightly changing expression of the appearance is called micro-expression.Because micro-expressions are human's spontaneous self-expression,they have extremely high value in judicial criminal investigation,public safety,physical therapy health and other fields.Therefore,micro-expression recognition has been paid more and more attention in the past decade.Micro-expressions have the characteristics of being short-lived and subtle changes in the face.At the same time,the appearance of micro-expressions is also accompanied by the offset of the face and the characteristics of environmental interference.How to use machine learning technology and models to recognize human facial micro-expression is a great challenge.Currently,the wide application of deep learning plays a very important role in the field of image recognition.Therefore,this study uses deep learning methods to cognize and recognize facial micro-expressions.The specific work includes the following aspects.Firstly,facial micro-expression recognition based on improved optical flow method and convolutional neural network.Considering that there is some redundancy in the micro-expression sequence fragments,this study uses only the onset frame and apex frame of each sample in the data set for identification.The optical flow method has a high advantage in extracting subtle dynamic features,but it has the disadvantages of large calculation and susceptibility to light.In order to solve this problem,an improved optical flow method is proposed.The optical flow field is obtained by using a lightweight convolutional neural network,and the time-series features of micro-expressions are extracted.At the same time,to solve the problems of low recognition rate and pre-complexity of traditional image recognition methods,using the advantages of deep learning in image recognition,the convolutional neural network is applied to micro-expression recognition to extract spatial information for micro-expression recognition.In the recognition of basic facial micro-expressions,it has a better classification effect than traditional classification methods.Secondly,use transfer learning methods to recognize facial micro-expressions.However,with the replacement of existing micro-expression samples,if the deep network is directly trained on the data set,over-fitting problems are likely to occur and affect the recognition performance.To solve this problem,transfer learning and deep learning are combined to enhance the learning ability of the network,accelerate the training process of deep convolutional neural networks,and improve the recognition ability.Through comparative experiments,the effectiveness of transfer learning in micro-expression recognition was verified.Finally,the design and implementation of a facial micro-expression recognition prototype system.Based on the theoretical research of this topic,a prototype system of facial microexpression recognition based on deep learning is designed and implemented.The system uses object-oriented ideas to show the implementation process and results of the algorithm in this study,and verifies the effectiveness and feasibility of the algorithm in facial micro-expression recognition.In this thesis,there are 41 figures,13 tables and 80 references.