A Face Verification System For Mobile Terminals Based On Deep Neural Network

With the development of high performance computing for mobile devices,the application of deep learning model on mobile devices has become a hot issue.Convolution neural network is a kind of deep learning technology.It can extract features of facial expression and end to end processing,it greatly improves the accuracy of image classification.Face verification is a kind of behavior that verifies the identity of the user by judging the similarity of two face images.As a new authentication technology,face verification is widely used in access control,attendance and other systems.Starting from the needs of identity authentication of truckers in logistics companies,this paper studies the application of deep convolution neural network model in mobile terminals,and proposes a mobile device based face verification system.We studied the main process of face verification,including face location,face alignment and face feature extraction.Then,based on the application of deep convolutional neural network in face feature representation,we studied the feasibility of applying deep convolution neural network model in mobile devices.In the implementation,through the model of space compression on the streamlining of the convolutional neural network model,the use of parallel computing to accelerate the face image prior to the efficiency of communication,so that the extraction can successfully run on mobile devices face feature model based on convolutional neural network.In addition,we also studied the impact of attribute prediction on the accuracy of face verification,and proposed a prediction algorithm based on deep convolution neural network.The experiment on the FERET database shows that the use of attribute prediction method can reduce the false detection rate of face verification to a certain extent.Finally,we implemented the face verification system based on Android system.The system can be deployed off-line in the mobile terminal devices equipped with Android system.It can get face images by camera,and do image processing in local area to complete face verification.In the algorithm,we used a deep convolution neural network to carry out image processing and face feature vector extraction to improve the accuracy of face verification.In implementation,we improved the efficiency of the algorithm by compiling JAVA and C++ code to adapt to the application of the deep learning algorithm on the mobile end.The experiment shows that the system can quickly complete the face verification process on the premise of ensuring the accuracy of up to 97.16%,and basically meet the needs of industrial application.