Micro-expression Recognition Based On Deep Neural Network

Micro-expression is a brief facial movement which reveals true feelings that a person want to conceal.Compared with the normal facial expression,micro-expression is more authentic.Of course,Micro-expression can be used as an effective basis for the identification of lies.Because of the short duration and low intensity of micro-expression,it is particularly difficult to be detected and recognized.In order to solve the shortcomings of traditional image recognition methods,such as low recognition rate and complex preprocessing of the input data.This thesis constructs a deep neural network to recognize micro-expression.The main work of this thesis is as follow:1.In this thesis,a convolutional neural network model is used to recognize microexpression.This model combines feature extraction and Classification in a unified framework.Compared with the feature extracted by normal manual methods,the feature extracted by convolutional neural network model shows better performance in generalization.This is important for the classification of micro-expressions.At the same time,this thesis also combines some methods such as dropout,data augmentation to further enhance the reliability of the model.2.After obtaining the convolutional neural network model which is used to recognize micro-expressions,this model can be used to extract the static feature of micro-expression.So this thesis present a method combining long short-term memory recurrent neural network model and convolutional neural network model to recognize micro-expressions.This method makes up the lack of dynamic characteristics of the features extracted by the convolutional neural network model,which can make full use of the information of the features in the time domain.The experimental results show that this method can recognize five kinds of microexpressions and obtain higher recognition rate than traditional methods in the CASMEII dataset.