Research And Implementation Of Passport Identity Verification Algorithm Based On Convolutional Neural Network

A passport is a common ID card issued by the government of a country or region to citizens of the country or region.Passport inspection refers to the verification of personal information of passport holders during entry and exit some customs clearance application scenarios to determine their legal identity.With the development of globalization,the use of passports for automatic identity verification has gradually increased,and the application prospects are more extensive.At the same time,it also faces the security risks of low recognition accuracy and fraudulent face attacks based on passport-based face verification methods.First of all,in view of the fact that the automatic passport inspection system is vulnerable to deceptive face attacks such as photos and videos in practical applications,this thesis proposes a face liveness detection algorithm based on depth maps,which uses deep network models to learn input face depth images converted from face point cloud data collected by 3D camera.Compared with the traditional LBP-based features,the spatial structure characteristics of the model have higher recognition accuracy and better performance.Secondly,in view of the problem that the text watermark in passports issued by some countries will obscure the face area on the passport page and affect the subsequent face verification,this thesis proposes a passport image text watermark removal algorithm,the network model is trained by continuously optimizing the average absolute error to obtain clean face images with occlusion watermarks removed.Finally,in order to solve the problem of large differences in the features of similar face samples in similar face verification scenarios and similar features of different types of face samples,it is difficult to accurately identify face verification.This thesis proposes a face verification algorithm based on improved Arc Face loss.Makes the network model more focused on face samples that are difficult to classify correctly during training,increases the loss contribution of difficult samples in the face training set,and solves the problem of class imbalance in the training process.The discriminative features learned through training can increase The inter-class feature distance also makes the intra-class features more compact,which further improves the accuracy of face verification.In this thesis,we conducted in-depth research on key algorithms such as face detection and face verification in automatic passport verification,and implemented a passport-based face verification system with strong security to ensure on-site image acquisition through live face detection The authenticity of the face in the face is used to verify the legal identity of the passport holder through face verification,and the validity and feasibility of the passport face verification method proposed in this thesis are verified through testing.