A Person And ID Verification System Based On Convolutional Neural Network

With the development of the real-name system,hotels,Internet cafes,banks and other industries have increased the need to verify the identity information of visitors.Departments that traditionally need to strictly verify personnel information,such as transportation and public security,also need to further develop the means of person-to-person verification.The traditional way of training face classifiers by comparing ID cards with ID card readers and manual tagging features is not suitable for mobile deployment and unlimited scenarios.In recent years,with the continuous development of computer hardware,the computing power of mobile phones is getting higher and higher,and the resolution of mobile camera is becoming clearer and clearer.And the continuous progress of in-depth learning also improves the ability of face verification in complex environments.This paper designs and implements a person and ID verification system based on convolution neural network.The system is built and operated on Android mobile phone.It uses mobile camera to obtain ID card and holder image,extract ID card number and face information,and verify ID card information and face verification.Three main tasks have been accomplished:Firstly,an identification algorithm based on convolutional neural network is studied and implemented.According to the relative position of ID card number,the area image of ID card number is intercepted during the shooting process;then the single channel extraction,edge information collection and number contour extraction are carried out according to the color and texture information of ID card;noise processing and image correction are carried out according to the inclined state of the noise and ID card in the shooting process;and then the image is cut by vertical projection method.Finally,the improved LeNet-5 convolution neural network is used to train and recognize the ID number.Secondly,face verification algorithm based on convolutional neural network is studied and implemented.The algorithm uses the directional gradient histogram method to obtain the face of the identity witness and the face of the holder;then the face is aligned by affine transformation;secondly,by adjusting the NN1 convolution neural network reasonably,optimizing the network structure and parameters to improve accuracy and real-time performance,a set of face coding vectors will be generated by training the aligned face images;finally,the similarity between two groups of face feature vectors is calculated for face verification.Finally,a prototype system of person-to-person verification based on convolutional neural network is implemented on mobile phones.Experiments show that the success rate of ID number recognition algorithm reaches 99.1%.It only takes 0.156 seconds to realize ID number recognition once;it takes only 0.72 seconds to realize the face verification of single ID witness and holder,and the recognition success rate reaches 99.71%.The whole process of ID card verification does not exceed 2 second,which fully meets the requirements of robustness,real-time and stability of ID verification system in practical use.