Research On Face Expression Recognition Algorithms Based On Mixed Features

With the advent of the era of artificial intelligence,intelligent human-computer interaction has become an important topic,among which emotional computing is essential.Emotional computing includes facial expression recognition,speech recognition and human behavior and posture recognition.As an important part of the field of emotional computing and human-computer interaction,facial expression recognition has become a challenging topic in the field of pattern recognition,attracting more and more researchers to devote themselves to the research in this field.Through reading and summarizing the existing literature on facial expression recognition,aiming at the shortcomings of traditional methods,this paper proposes a facial expression recognition algorithm based on hybrid features,taking static facial expression images as the research object.The main work of this paper is as follows:(1)Firstly,the original image is preprocessed,grayed,histogram equalization,face detection,scale normalized and facial feature landmark location.At the same time,in order to verify the reliability of the algorithm,a facial expression database based on 10 human models is built.(2)Aiming at the problems of traditional face feature point location methods,such as active shape model(ASM),which are sensitive to image noise and easy to fall into local minimum because of the unsatisfactory location of initial feature points,this paper uses an Ensemble of Regression Trees to locate face feature landmarks.Through gradient lifting tree method,a large number of simple regression trees are combined into strong regressors,and face feature points are accurately located by iteration.Experiments show that compared with ASM,the recognition rate of extracting geometric features is higher when facial feature points are located by this method.(3)In the aspect of Gabor feature extraction,aiming at the shortcomings of traditional feature extraction,this paper divides face,eyes,nose and mouth based on feature point location,removes redundant information,and selects the parameters of Gabor filter closest to the receptive field characteristics of visual cells to extract features,which has a higher recognition rate than traditional Gabor feature.In the aspect of LBP feature extraction,uniform LBP mode is selected and histogram features are extracted in blocks to ensure the balance of local and global information.In the aspect of geometric feature extraction,this paper proposes a joint geometric feature.On the basis of feature point location,the distancefeature and deformation feature are extracted,and the joint geometric feature is formed.Compared with the traditional geometric feature,the recognition rate is higher.(4)A facial expression recognition method based on three feature mixtures is proposed in this paper.Firstly,the integrated regression tree model is used to locate feature points and extract joint geometric features;secondly,important regions of eyes,nose and mouth are divided to extract and optimize Gabor features;secondly,LBP features are extracted by histogram,and then three support vector machine(SVM)classifiers are trained with these three features respectively.Then,the weighted voting method proposed in this paper is used to make a comprehensive decision on the results of the three classifiers,and the final recognition results are obtained.The experimental results show that the algorithm has higher recognition rate and robustness than single feature,and gives full play to the advantages and characteristics of different features...