Facial Expression Recognition Based On Convolutional Neural Network

Facial expression recognition is one of the research hotspots in computer vision field.Because facial expression recognition has important practical significance in social robotics,medical treatment,driver fatigue monitoring and many other human-computer interaction systems.The research about facial expression recognition is becoming hotter and hotter.Researchers have proposed many valuable methods based on traditional machine learning in the field of facial expression recognition.However,the advantages of deep learning methods in the field of facial expression recognition have become increasingly prominent,many methods based deep learning have been proposed.On the basis of summarizing and analyzing existing facial expression recognition algorithms,this paper proposes two methods by combining traditional machine learning and deep learning.And two methods have achieved good accuracy.Main work:(1)This paper introduces the research background and application prospect in the field of facial expression recognition.And this paper summarizes the research status at home and abroad and introduces the related basic knowledge.(2)Three networks were built in this paper.Three networks are VGG13,VGG16 and VGG19.Because LBP operator have the advantages of gray invariance and strong resistance to background noise and visible light,LBP feature extraction and data enhancement were applied to these networks.And these networks were verified on CK+ dataset.The experimental results show that the accuracy of the network using LBP feature extraction and data enhancement is9.1% higher than network without LBP feature extraction and data enhancement.(3)Ensemble learning can combine several weak supervision models to get a better and more comprehensive strong supervision model.Even if one of the weak classifiers gets the wrong prediction,other weak classifiers can correct the error.Three networks were built in this paper.Three networks are VGG19,Resnet50 and Resnext50.The Bagging ensemble learning algorithm and data enhancement are applied to these networks.And these networks were verified on the FER2013 dataset.The experimental results show that the accuracy of the network using Bagging ensemble learning algorithm and data enhancement is 12.2% higher than network without Bagging ensemble learning algorithm and data enhancement.