The Research On Facial Expression Recognition Algorithm Based On Convolutional Neural Network

In recent years,facial expression recognition technology has been widely used in human-computer interaction,computer vision,security monitoring and image classification,etc.But in practical application,the collected facial expression images are easily affected by the factors such as imaging angle and acquisition equipment and other factors,which makes the expression image appear in the plane rotation.It leads to the decline of facial expression recognition rate,which is difficult to meet the actual needs.Local Binary Pattern is a kind of feature extraction algorithm with rotation invariance,which can solve the problem to a certain extent.However,when LBP operator is used to extract expression features,it is necessary to set the block size and quantity of the image artificially.As a result,some facial expression information is lost.To solve the above problems,a facial expression recognition algorithm based on Convolution Neural Network and Local Binary Pattern is proposed in this thesis.The main contents of this thesis are as follows:1.When the classical Alex Net Convolutional Neural Network is applied to face expression recognition,there is a problem of low expression recognition rate.In this thesis,a facial expression recognition algorithm based on an improved Convolution Neural Network is proposed.Using the idea of continuous convolution,a new Expression Net Convolutional Neural Network was constructed by modifying the Alex Net model.Among them,the Expression Net model is made up of the structure unit that has two continuous convolutional layers and a max pooling layer.From the aspects of performance,this thesis compared the proposed algorithm with Single Net model which has three convolutional layers and Alex Net model.Meanwhile,the proposed algorithm was compared with multi-column Convolutional Neural Networks and Convolutional Neural Network with inception structure,etc.The experimental results show that compared with the Single Net model and the Alex Net model,the algorithm can improve the expression recognition rate to a certain extent.At the same time,the algorithm in this thesis has a certain robustness compared with other Convolutional Neural Networks.2.Aiming at the lack of rotational invariance in Convolutional Neural Network,Local Binary Pattern algorithm that is robust to rotation is introduced.A facialexpression recognition algorithm based on CNN features and LBP feature fusion is proposed.Firstly,Expression Net model was used to extract the convolution features of expression images.In the meantime,LBP operator was used to extract expression features with rotation invariance.Secondly,the features of CNN and LBP were serially fused in the second full-connection layer of CNN,and dimension reduction was performed by principal component analysis.Finally,the processed expression features were input into the softmax classification layer,and the categories corresponding to the neuron with the largest output value were selected as the final classification result.The simulation results show that the algorithm can improve the expression recognition rate under the rotation condition to some extent.