Research On Face Recognition System Based On Deep Learning

Computer vision technology has gradually developed since 1960 s.Due to its noninvasiveness,convenience,friendliness,non-contact and scalability,face recognition has received more attention.After 2012,big data and deep learning have developed rapidly.There are many new face recognition algorithms that based on deep learning to appear,and their effect are much higher than the traditional method.A real-time face recognition system based on deep learning technology has great significance and application prospects.This paper focuses on the design and implementation of a face recognition system based on deep learning technology.The specific research contents and innovations are summarized as follows:(1)In the module of face detection,the development and application of deep learning in the field of face detection are introduced.The cascaded convolutional neural network and the non-cascaded convolutional neural network are discussed,and trained on the CIFAR dataset to compare the detection accuracy and processing speed.MTCNN(Multi-task Cascaded Convolutional Network)is selected to finish the task of face detection.(2)In the module of face recognition,a new analysis tool is proposed.This tool uses the SE-Block(Squeeze-and-Excitation Block)to obtain the weights of each level of features,and then analyze the structural characteristics of the network.A new network structure called Dense-Residual Network(DRNet)is proposed.The shallow part of the network uses densely connected blocks to enhance the ability of mining new features.The deep part of the network uses residual blocks to improve training performance.The network reduces the parameters while improving the ability to express features.A FaceNet model based on DRNet is designed to finish the task of face recognition.At the end of the paper,the system function was tested.(3)This paper designs a face recognition system based on deep learning technology.The system obtains video through cameras,uses MTCNN to perform face detection on each frame of videos,crops out the region of faces,extracts the deep features through DRNet,and maps the features into 128-D feature vectors through the spherical mapping method.In the end,the system achieved an accuracy of 98.35% on the LFW data set.Deep learning technology is very suitable for face recognition.The face recognition system developed based on this technology with the support of appropriate hardware can achieve higher recognition accuracy and real-time performance.It can fulfil requirements of practical applications.