Research On Expression Recognition Based On Deep Neural Network

As an important means of expressing personal emotion in human communication,facial expression plays a crucial role in human communication.At the same time,facial expression is also the premise for computers to understand human behavior and the basis of emotional computing.It has a broad application prospect in human-computer interaction,medical treatment,public safety and other fields.Facial expression recognition usually consists of two steps: facial feature extraction and classifier selection.How to extract effective facial features is still a challenging research topic.Traditional facial expression recognition methods require manual extraction of features,which is tedious and unstable.In recent years,with the rapid development of deep learning technology,deep neural networks have achieved outstanding performance in the field of computer vision.It can automatically extract more abstract and more accurate high-dimensional features through learning layer by layer,and has stronger generalization ability.Therefore,this paper studies the current popular deep neural network,analyzes its structure and applies it to the field of facial expression recognition.The main work of this paper is as follows:(1)An expression recognition method combining image difference and convolutional deep belief network is proposed.When extracting facial features,traditional recognition methods are difficult to remove individual differences and extract high-level features of key parts of facial expressions.The introduction of image difference can highlight the information of the key parts of the expression images,which can be used as the input of the convolutional deep belief network to extract more accurate expression features,and the difference images of various expressions can be extracted by the difference operation of various expression images and neutral expression images.After that,the traditional convolutional restricted boltzmann machine is improved,the visible layer is divided into multiple regions,and the features are learned by partitioning.The improved CRBMs are stacked to form a partitioned convolution deep belief network,which is trained by average contrastive divergence algorithm layer by layer and the softmax network is added to the top layer as the output layer to form the classification surface;Finally,the partitioned convolutional deep belief network which is well-trained is used to identify the expression images from test samples.(2)An expression recognition method based on ensemble convolutional neural network is proposed.A single neural network is usually not comprehensive enough for facial expression recognition,and the effect of facial expression image recognition is limited.Using ensemble learning method to integrate multiple neural networks into one system can make neural networks of different structures play a synergistic role in facial expression recognition.Based on the unique advantages of convolutional neural network in image classification,the performance of convolutional neural network in facial expression recognition is discussed.First,three convolution neural networks with different structures are constructed.After that,the three neural networks are fused by ensemble learning to achieve robust and accurate facial expression recognition.We assign different weights to the three neural networks to measure their contribution to the final recognition results.Finally,the trained integrated neural network is used to identify the test samples.Experimental results on two classic expression datasets show that,this method can extract more accurate and effective expression features to achieve expression recognition compared with the traditional single deep neural network and some methods that do not use deep learning model.In this paper,experiments use the facial expression recognition method based on the deep neural network on JAFFE and CK + datasets show that the deep neural network can extract effective features,and recognition effect is superior to the traditional method.Moreover,the method combining ensemble learning and deep neural networks is better than the single network in recognizing different expressions,and it can achieve better classification effect through synergistic effect in feature extraction task,which provides help for further research on the application of deep neural networks in the field of facial expression recognition.