Research On Face And Expression Recognition Based On Convolutional Neural Network

In the past two decades,biometric recognition has been a research hotspot in the field of machine learning and pattern recognition.Human face is the most intuitive biological signal of human beings,and the research of biometric recognition based on face image has important application value.With the maturity of deep learning methods,convolutional neural networks are a popular method in deep learning.Convolutional neural networks are more and more widely used in the field of image processing.In this regard,this paper studies face and expression recognition based on convolutional neural networks.The main research work completed in this paper is as follows:Firstly,According to the current development trend and subject requirements of face and expression recognition,on the basis of mastering the relevant algorithms of convolutional neural network,an overall design scheme for face and expression recognition is proposed.Secondly,the problem of how to improve the accuracy of face recognition is studied.The original Le Net-5 network is improved by changing the number of network layers and the size of convolutional layers,increasing the number of convolutional layers,using Gabor filter initialization and using PRe LU activation function to improve,and proposes a new model The experimental results of Gabor-Le Net and Gabor-Le Net network on the three face data sets of ORL,GT and AR show that the accuracy of face recognition is increased to 97.50%,96.79% and97.58%,respectively,compared with the original Le Net-5,The accuracy of face recognition on the network has been greatly improved.This shows that the model proposed in this paper can achieve a good face recognition effect.Thirdly,aiming at the problem of large parameters of convolutional neural network and low accuracy of traditional expression recognition methods,this paper proposes an expression recognition model based on an improved Dense Net model.On the basis of the dense convolutional neural network Dense Net,the LLU-Dense Net model is proposed by adding dense blocks in the network layer,using the convolutional layer Gabor filter initialization,and using the new activation function LLU(x).The feature reuse and parameter compression technology used in this model improves the learning ability of the network and greatly reduces the model parameters.By comparing the experiments with the Res Net18 and Alex Net networks,the LLU-Dense Net model has achieved 99.78%,70.78% and 85.43% accuracy in the experiments on the CK+,FER2013 and FERPlus three data sets,which are better than those of the previous two networks.The effect is better,which shows that the LLU-Dense Net model has achieved better results in facial expression recognition.At the same time,a facial expression recognition system is designed based on this model,which can perform image recognition and online video recognition.