Research On Facial Expression Recognition Based On Deep Learning

In recent decades,human-computer interaction and computer vision have been an important area of computer research,and direct communication between computers and people has been a concern.A lot of research has been done to improve and develop human-computer interaction.An important factor that promotes the enhancement and development of the interaction between computers and humans is to study the ability of computers to recognize human facial expressions.Facial expression plays an important role in people's daily communication.In recent years,deep learning has made great progress in image recognition.After studying a large amount of literature,this paper decided to apply deep learning to facial expression recognition.However,deep learning takes too long to train data sets,and it is easy to fall into over fitting when the number of data sets is small.Due to the small number of pictures in the current facial expression data set,this paper proposes a series of network frameworks to overcome the shortcomings of deep learning in facial expression recognition.The work of this paper is summarized as follows:1.When a single-channel convolutional neural network model recognizes images,the size of the obtained feature map is single,which will cause the network to fail to obtain more feature information of the pictures.In order to overcome this shortcoming,this paper proposes a double-channel convolutional neural network model.The model has two parallel sub-channels,and each channel is a convolutional neural network.In this way,the network can obtain two feature maps of different sizes,which is helpful for the network to obtain more information of the pictures.2.In the full connection layer,maxout activation function is combined with dropout,which makes the anti over fitting ability of the network stronger.3.An influence factor ? is proposed.A double-channel convolutional neural network has two channels,but each channel is affected differently by the fully connected layer during backpropagation.Therefore,the backpropagation error of the fully connected layer is assigned to the two channels.? Can be used to adjust the proportion of distribution and improve network performance4.A CNN expression recognition model based on ASM algorithm was proposed.In this algorithm,we first use the ASM algorithm to mark 68 feature points on the face,then use the coordinate information of these feature points to cut the faces in the picture,and finally send the cut face pictures into Facial recognition in a double-channel convolutional neural network.This algorithm not only improves the accuracy of expression recognition,but also accelerates the convergence speed.We identify and judge the algorithm proposed in this paper on the data sets of Jaffe,CK + and Fer2013,and compare it with ResNet and VGG-16.The results show that the algorithm proposed in this paper has higher recognition accuracy and faster convergence speed compared with the other two models,which verifies the effectiveness of this method.