Design And Implementation Of Identity Document Verification System

The identity document verification system can verify the identity of the document and the holder in different scenario such as airports and railway stations.The intelligent document verification system based on computer vision techniques can improve the efficiency and reliability of identity verification.In the intelligent document verification system,complex backgrounds in the image will interfere with the document region location,and affect the automatic reading of document information.In addition,the existing face verification algorithms can hardly handle the face images with different resolutions in various documents.We focused on solving the problem of document region location and multi-resolution face verification in our system,then designed and implemented the identity document verification system.Firstly,aiming at document region location in complex backgrounds,we proposed a document region location method combining convolutional neural network and Canny operator.To reduce the interference of background noise,we used a convolutional neural network which fully utilized the multi-scale features and high-level semantic information in image to extract edges,and merged the edges generated by the convolutional neural network and the Canny operator to improve the fineness of the edges.Based on this,a series of rules were set to eliminate the interference of non-document edges,and improved the accuracy of document region location in the output edge image.Secondly,for the problem of multiresolution face verification,we proposed a face verification algorithm based on a twobranch convolutional neural network,by which the facial features are extracted from highresolution and low-resolution images,respectively.For the low-resolution facial features extraction network,it minimizes the distance of different resolution face images of the same person in the feature space as the training target,so as to improve the accuracy of the reconstructed face images in the verification tasks.In addition,we used a joint Bayesian classifier to classify the extracted facial features to improve the accuracy of face verification.Finally,we designed and implemented an identity document verification system with graphical user interface,which includes functions such as document information collection,identity document verification and document information database.On the document images dataset,the document region location algorithm we proposed yielded better performance than the typical methods in mean Intersection-over-Union.On the dataset of face images with different resolutions,the multi-resolution face verification algorithm we proposed achieved higher verification accuracy.Besides,the document verification system we designed has been applied to the self-service exit-entry in Beijing Daxing International Airport,which demonstrated the feasibility and effectiveness of the proposed methods.