Design Of Face Recognition System Based On Convolutional Neural Network

Recent years have witnessed a spurt of progress in AI technology which helps to bring the facial recognition technology to the fore in computer science field.Integrated with the deep learning algorithm,facial recognition has realized leapfrog development in its efficiency.Now facial recognition has brought benefits to almost every aspect in our life,among which smart phones stand out as the pioneers of this technology.Subsequently,facial recognition technology has been widely used on the Internet,which accordingly begets expression recognition,human body pose recognition,etc.And as a reliable identification recognition technology,it plays an essential role in security field.In addition,it has greatly facilitated people's life in terms of such areas as high-speed rail ID check-in,access control and clock in.therefore,to process flood of data,the thesis introduces the deep learning technology and creates,by the means of framework Caffe,a convolutional neural network(CNN)to conduct facial recognition.On Windows and Linux platforms respectively,the client and server of facial recognition have been designed to establish a complete facial recognition system by using the TCP/IP protocol.Firstly,the thesis introduces the current condition and application fields of CNN and facial recognition as well as the deep learning framework Caffe,of which the strengths,three modules,basic structure and development environment will be expounded in comparison with other deep learning framework.Then,a CNN with 5 convolutional layers and 3 pooling layers is designed,and the face models are trained by using the Caltech face database.Through feature visualization analysis and experimental results,a conclusion can be drawn that the face model designed in the thesis can better display facial features and has higher recognition rate compared with the traditional methods.Besides,compared that with such mature convolutional neural networks as VGG,the network model in the paper,fewer-layered and less complex,can perfectly meet the requirements of.Next,the trained face models will be applied on the Windows platform to interact with users,collect facial images,and communicate with the server.The facial recognition server implemented on the Linux platform ensure the implementation of functions including the registration,login and uploading of facial information as well as the interaction with the My Sql date base for storage.Finally,tests carried out upon the designed facial recognition system have led to a truth that in the case of enough light and no face occlusion,the system,with high accuracy rate,can meet daily product needs in different systems,namely,educational administration,staff management,access control,etc.