| Facial expression is an indispensable way of human communication.Through the study of facial expressions,human psychology can be explored,and the behavior of people’s intentions can be fully understood.Deep learning is a feature learning method.It transforms the data into higher-level and more abstract expressions through simple non-linear models,and it can solve the voice processing,computer vision,natural language processing and other issues.In this paper,deep learning is used to solve some of the problems in facial expression recognition,and the result is verified by experiment.The main research content of this article is as follows:1.Many types of deep learning models are studied in this paper.According to the structure it can be divided into deep convolution neural network,deep belief network,deep boltzmann machine,stack automatic encoder and recursive neural network.They have different algorithms,and the applicable areas are not the same.Therefore,choosing the appropriate deep learning model is the key to solve the problem of facial expression recognition.In this paper,the deep convolution neural network was selected as the deep learning model.2.Aiming at the problem that the feature extraction in the static expression recognition will lose the original characteristic information of the image,the method based on deep convolution neural network is put forward in this paper to achieve the expression feature extraction.Because the deep convolution neural network avoids the complex pre-processing of the image,the original image can be input directly.It is extracted by the combination of convolution and pooling.It does not need artificial extraction feature,and the network is easy to train.The fully connected neural network has better generalization performance.So the deep convolution neural network is applied in static expression recognition.3.Aiming at the problem of poor anti-interference,speed and real-time performance in dynamic expression recognition,deep convolution neural network is used in this paper to achieve dynamic expression feature extraction.As the dynamic expression recognition system input real-time access to the dynamic facial expression sequence,which is different from the static expression recognition,it requires the system has real-time access to store and identify the faces.In order to solve this problem,the Haar classifier is used to detect the face,and then the deep convolution neural network is introduced.The essential features of the image are constructed and the expression features are extracted.The Softmax classifier is used to realize the expression classification.4.In order to improve the nonlinear expression ability of deep convolution neural network and achieve better expression feature extraction,a deep continuous convolution neural network is proposed for facial expression recognition.This paper refers to the idea of multi-layers small-scale convolution instead of single-layer large-scale convolution.It uses 2 layers of continuous convolution to replace the single layer convolution layer,to improve the network of non-linear expression,and adjust the activation function and parameter optimization method of the network to improve the facial feature extraction ability of the network. |
