Facial Expression Recognition Based On Multiple Features And Convolutional Neural Networks

With the development of computer technology,facial expression recognition has gradually become a hot research topic in the field of pattern recognition and artificial intelligence.Expression is the basis of emotional understanding,and it can reflect the psychological emotion of the identified object to a certain extent.Facial expression recognition enables computers and robots have the ability to understand and express emotions,so as to achieve better human-machine interaction.It has applications in psychology,intelligent robots,safe driving and other fields.There are variety of facial expression feature extraction algorithms,but the recognition rate of facial expression is difficult to be improved due to the influence of light,attitude,age,occlusion,and the limited number of samples in the expression database.In order to improve the recognition rate of facial expression,this thesis studies the algorithm of facial expression recognition,and proposes a facial expression recognition method based on feature fusion and convolutional neural network.The main work is as follows:First,a series of preprocessing operations are performed on facial expression images.Expression images are transformed from RGB space to gray space by the weighted average method.Expression images are cropped by using the model of “three court five eyes” and the geometric model of the face,scaled by bicubic interpolation.The histogram equalization is applied to enhance the local contrast of the image.Secondly,the facial expression feature extraction algorithms are studied.The characteristics of LBP,Gabor wavelet transform and convolution layer are visualized and analyzed.Then,this thesis introduces the proposed facial expression recognition algorithm based on multi-feature and convolutional neural network.Three feature maps of DWT low-frequency sub-band,LDP feature and Sobel feature are extracted respectively..An eight-layer convolutional neural network model FERNet(Ficial Expression Recognition Networks)is designed by drawing lessons from the classic LeNet-5 and AlexNet model.Relu is used as an activation function and the Dropout layer is added to prevent the overfitting in the course of network training.To trian the network model,three feature maps are used as the input of FERNet.After applicating PCA in dimension reduction of feature vectors extracted from the convolutional neural network,the expression feature vectors after dimension reduction are obtained.Finally,Support Vector Machine(SVM)is used to classify and recognize facial expressions.To verify the effectiveness of the proposed method,the CK+ and RAF database are used in the experiment,the recognition rates of facial expressions are 95.32% and 68.94% respectively.The results show that the proposed algorithm performs better than LBP,Gabor,GB-DBNs+SAE and BDBN algorithms.