Research On Facial Expression Recognition Methods Based On Deep Learning

With the development of computer vision and artificial intelligence technology,research on facial expression recognition has become a hot spot.Because changes in facial expressions play a powerful role in human-computer interaction and non-verbal communication,facial expression recognition has a milestone significance for the development of interactive computer systems.Facial expression contains rich behavior information.It is a signal that humans convey information.It is a difficulty to let machines fully learn expression features and improve recognition rate in current research.This paper studies facial expression recognition methods,the specific works are as follows:(1)Traditional expression recognition method is studied.The first is face detection.Face detection method based on skin color segmentation and face detection method based on Adaboost algorithm are introduced.Then two algorithms for expression feature extraction: LBP and HOG are studied.Propose a feature fusion expression recognition method for the defect that single feature describing images has insufficient recognition rate.The experimental results show that it is better to use fused features for facial expression classification.(2)Expression recognition method based on convolutional neural network is studied.Convolution neural network is one of the important models of deep learning.First,the working mechanism of convolutional neural network is introduced,including convolution operation,pooling operation,activation function,loss function and network training methods.Then,common open source frameworks for deep learning are introduced,and their characteristics are analyzed.Finally,a convolutional neural network with a few layers is designed to perform experiments on both CK+ and JAFFE facial expression datasets.(3)Expression recognition method of improved VGG-19 based on transfer learning is studied.Choose Multi-task Convolutional Neural Network to finish face detection,and experiment results indicate that MTCNN has a better detection effect than the method based on Adaboost algorithm.Aiming at the problem of low recognition rate caused by small scale of expression datasets,an improved VGG-19 expression recognition method based on transfer learning is proposed.The experimental results show that the method proposed in this paper has achieved higher recognition rate on both CK+ and JAFFE facial expression datasets.