Identification Algorithm And Research Of Identity Card Number Recognition

The second generation ID card is usually used to as proof of citizenship identity,and the identity card number is unique number,with the continuous development of optical character recognition technology,document identification technology is widely used to realize the identification and entry of number in ID card image will bring great convenience.In the past,the acquisition of the ID card image is carried out by the scanner,but it is not appropriate for real-time access to the image like some platforms require an ID card to upload an identity card.This paper is based on the camera to capture the identity card to obtain the image,using processing technology and deep learning technology to complete the number of image recognition.The main contents of this thesis are as follows,ID card image analysis,image preprocessing,character segmentation and character recognition,details as follows:1.The camera is used to photograph the ID card,and the image of the ID card is analyzed based on the captured image.Unlike the scanned ID card image,the image needs to be preprocessed to complete the character recognition.2.In the pretreatment using LSD and perspective transformation to complete the integration of the image,first use the straight line detection of the original ID card edge line,find the edge of the ID card,and finally find the edge of the four corners of the ID card,and then use perspective transformation to correct the image,get the image of the standard ID card.3.After obtaining the ID card image,according to the ID number in the ID card appears in the approximate position,the use of connected component labeling to extract the identity card number image;in order to get a single character need to identify the ID card image segmentation,a single gray scale image is obtained using improved projection segmentation methods.4.In order to identify a single number image,a method of convolutional neural network is applied to complete the recognition of the number image,it is not necessary to normalize the input data in advance and achieved good results.