Facial Expression Recognition Based On Shallow Convolutional Neural Network

With the development of society and science and technology,the way of infor-mation transmission has changed dramatically.In the process of human communica-tion,more than half of the information content is transmitted through images,so fa-cial expressions play an irreplaceable role in the process of interpersonal communica-tion.Thanks to the rapid development of computer vision,facial expression recognition(FER)has been widely used in human-computer interaction,online education,psycho-logical therapy and other fields in recent years.However,facial expression recognition is still a very challenging research topic due to the differences in brightness,background,posture and other aspects of the application site.Deep learning methods need the sup-port of a large amount of data.If the data is small,the learning ability of model will be weak and the model will not be ideal.Moreover,it is very difficult and time-consuming to train the deep neural network.Feature extraction is very important for FER,but the traditional image feature extraction method lose a lot of useful information…These are all challenges in the field of FER.Based on the previous research,this paper proposes many improved methods and presents a complete set of FER system flow.This paper proposes a shallow convo-lution neural network(CNN)based on the LeNet-5 for FER.After removes two fully connected layers,the network model is more simple and reduces the training and testing time,it makes that the convolutional neural network model can be trained within half an hour,and takes 5.51ms for prediction.With the aim of coping with few data and extracting only useful features from image,we propose new face cropping and rotation strategies and simplification of the CNN to make data more abundant and only useful facial features can be extracted.Finally,this paper also proposes a new face cropping method,which removes the useless information in the image,so that the image only retains the local area which is beneficial to the expression recognition and it greatly improves the recognition accuracy.Experiments to evaluate the proposed method were performed on the CK+and JAFFE databases.High average recognition accuracies of 97.38%and 97.18%were obtained for 7-class experiments on the CK+ and JAFFE databases,respectively.A study of the impact of each proposed data processing method and CNN simplification is also presented.The proposed method is competitive with ex-isting methods in terms of training time,testing time,and recognition accuracy.