Research On Expression Recognition Algorithm Based On Optimized Convolutional Neural Network

In recent years,the development of artificial intelligence and related disciplines is relatively rapid,people also gradually feel the joy and convenience of artificial intelligence for life,at the same time,the demand for human-computer interaction is increasing.The introduction of facial expression recognition technology will help to improve the anthropomorphism of human-computer interaction.It is found that the convolution neural network algorithm is more effective than other traditional methods in extracting image features in some fields,so many researchers have tried to apply the convolutional neural network to the field of facial expression recognition.However,convolution neural network also has some shortcomings that can not be ignored.Because of the small number of layers,the shallow convolution neural network has very limited ability to extract image features when the sample data is limited,and the sample data is also vulnerable to external factors such as illumination,size,occlusion and so on in the process of feature extraction.Therefore,this paper proposes an expression recognition algorithm based on improved VGG convolutional neural network and combining Gabor with convolutional neural network to solve the above problems.Deep convolutional neural network is based on a kind of deep nonlinear network structure to extract deeper features of images,showing a strong ability to learn features from images.In order to achieve the accurate classification of facial expression images under normal conditions,an improved facial expression recognition model based on VGG-19 convolutional neural network is proposed.Because most of the expression databases used for facial expression recognition lack enough data to train the whole network from scratch,this paper adopts the transfer learning technology to overcome the problem of insufficient number of images.By optimizing the network structure,the network parameters can be reduced properly to prevent over fitting in the training process.The method was applied to the CK+ expression database for facial expression data training and data analysis,and the accuracy of facial expression recognition finally reached 96%.However,the model of deep convolution neural network is too complex,whichwill lead to increased training difficulty.Under the condition of less training sample data and complex deep network model training,in order to improve the effect of expression recognition,it is essential to enhance the performance of feature extraction.For the sake of extract the deeper expression features in the image,and according to the advantages of Gabor,which can extract the texture features in the image well,this paper proposes an expression recognition algorithm combining Gabor and convolution neural network.Before the image is input into convolution neural network,this algorithm first extracts the Gabor features of the expression image,and adds Gabor As the input of convolution neural network,the feature graph improves the expression power of the algorithm.The optimal expression recognition rate obtained by using this algorithm can reach 99.3%,which proves that this method can effectively extract facial features.