Research On Emotion Recognition Method Based On Facial Expression

In today's intelligent era,emotion recognition based on facial expression has high application value in human-computer interaction and other fields.If the researchers can recognize the basic expression and the degree of expression through the intensity information of facial expression,it will be helpful for the computer to recognize and analyze the deeper emotions of human beings.Therefore,this thesis takes emotion recognition as the main research content,and studies the recognition method of facial expression degree on the basis of recognizing the basic facial expression.Basic facial expression recognition divides facial expressions into seven basic expressions,such as happy,sad,fear,anger,disgust,surprise and neutral.Based on the recognition of expression degree,each basic expression is divided into five levels according to the change degree of its facial expression.In this thesis,the specific implementation method of basic facial expression recognition is described in detail.It includes three basic steps: facial expression image preprocessing,facial expression feature extraction and expression classification and recognition.The key step is facial expression feature extraction and facial expression classification recognition.In this thesis,two facial expression recognition algorithms based on LBP + SVM and HOG + SVM are adopted respectively.The recognition rate and running time of the two algorithms are compared and analyzed.It is found that the algorithm based on HOG + SVM is slightly better than the algorithm based on LBP + SVM.Therefore,this thesis selects the algorithm based on HOG +SVM as the basic facial expression recognition method.The recognition rate of this method on JAFFE and CK + data sets can reach 92.89% and 93.45% respectively,which can meet the needs of subsequent facial expression recognition.For facial expression degree recognition,this thesis proposes a facial expression degree recognition algorithm based on the distance variation of feature points,which includes the following three steps: facial feature point location,facial expression feature extraction and facial expression degree classification and recognition.Firstly,face feature points are located,and 68 feature points are selected in the input facial expression image using dlib database model.The second is facial expression feature extraction.This thesis proposes a feature of distance variation of facial feature points.The feature takes the distance variation between some facial feature points and the reference point in the process of facial expression change as the feature value of expression degree to reflect the change degree of facial expression.In order to improve the recognition efficiency of the algorithm,it is necessary to select the corresponding distance change features of 68 feature points according to the change features of various expressions.In this thesis,the experimental analysis method is used to select the feature values that can accurately reflect the basic expression degree for each expression.Finally,the feature vector is constructed by using the selected distance variation to classify and recognize the expression degree.BP neural network classifier and SVM classifier are used to test the recognition rate of the proposed method based on the distance variation feature.The recognition rate of each basic expression can be more than 80%.Compared with the existing facial expression classification algorithm,the recognition rate is higher,which proves that the performance of the proposed method is great.