Research On Face Recognition And Facial Expression Recognition Based On Convolutional Neural Network

With the development of computer vision and intelligent robots,face recognition and facial expression recognition have become a research focus.Traditional methods often require manual selection of features,we often fail to obtain an effective face features completely or correctly,and they are susceptible to interference from environmental noise.In recent years,deep learning algorithms are often used in computer vision tasks,which has made great progress in pattern recognition technology.Based on convolutional neural network,this paper mainly constructs a face recognition system,and also studies and implements the facial expression recognition algorithm.The main research contents of the thesis include:(1)Research on face detection.The purpose of face detection is to reduce the influence of environmental noise as much as possible,and it is also the first stage of face recognition and facial expression recognition tasks.This paper studies and implements the traditional methods and the deep learning method.Based on the multi-task cascade convolution neural network(MTCNN),this paper realizes a face detection algorithm with high accuracy and robustness.(2)Research on face recognition systems.The face recognition system built in this paper has the function of liveness detection,which can effectively prevent false faces and other forged face attacks,the system has strong system reliability.Because of the convolution neural network method,the system has good feature extraction and generalization ability to face features.It can adapt to uneven illumination,facial expression,distance and even a certain degree of occlusion.It can accurately identify the target identity in single and multi-person mode.(3)Design of the Xception-small neural network model.Based on the research on the principle and characteristics of convolutional neural networks,this paper proposes an efficient convolutional neural network model for facial expression recognition tasks.The depthwise separable convolution module structure and batch normalization regularization methods are used,and the global average pooling layer is used to replace the full connection layer,and the small convolution core is used to replace the large convolution core as far as possible.These techniques enhance the network expression ability,reduce the number of model parameters,improve the accuracy and shorten the training iteration time.(4)Research on facial expression recognition algorithm.Firstly,the CK+expression dataset is preprocessed and augmented.Based on the convolutional neural network model proposed in this paper,contrastive experiments are conducted for different hyper-parameters and optimization methods,so that the model can be optimized.By visualizing the training process and results,the training effect of the model can be evaluated intuitively.This paper also evaluates the performance of the algorithm,compares the performance of CK+dataset with traditional methods and other deep learning methods,and carries out experiments in natural scenes to verify the generalization ability of the algorithm and the accuracy of facial expression recognition.(5)Implementation of human-computer expression interaction.Based on NAO,the human-computer emotional interaction function is realized,which enables the robot to recognize facial expressions and perform different voice and motion interactions for different expression categories.