Design And Implementation Of ID Card Information Extraction Based On Neural Network

With the rapid development of the Internet and the gradual improvement of information technology,the requirements for text recognition in our lives and production are getting higher and higher.The ID card is a kind of document that proves the identity of the holder and is the identity certificate of each Chinese.The biggest advantage of ID card automatic recognition technology over manual extraction is fast,accurate and safe.Nowadays,many institutions or companies are not authorized by the public security organs.The method of extracting the information of the ID card is extracted manually.This process not only consumes a lot of manpower,but also is not very high in accuracy.Although some high-end ID card identification devices are used in public security departments,hotels,Internet cafes,etc,the price required for the purchase of the instrument is very expensive,so the application of the instrument in life is not particularly common,but the ID card The demand for text recognition technology in our life and production is very large,so we need to produce an affordable and efficient ID card recognition instrument,so that the instrument can be applied to our life.The school records information of the students,the company records the information on the paper form,etc,can use this ID card text recognition system,and can obtain very accurate results on the basis of low purchase price,which is very huge value in the future production and life.The text information of China's ID card includes: name,gender,ethnicity,date of birth,address and citizenship number.This paper studies the information extraction of ID card based on neural network,including the preprocessing of ID card image,the extraction and screening of text area and the character recognition of text,and based on the research content of the thesis,an efficient and fast ID card text information extraction system is designed.(1)Firstly,in order to improve the recognition efficiency of the entire ID card text information extraction system,the ID card image is scaled and grayscaled to reduce the image pixel size and channel number.In order to improve the accuracy of subsequent text recognition,it is necessary to correct the image of the ID card.The tilt correction process can be divided into rough correction based on neural network and precise correction based on vertical projection algorithm,which makes the correction result more accurate and improves the accuracy of the text region extraction.(2)Before the text character recognition,the text area in the ID card image needs to be extracted.In this paper,a new type of target detection algorithm is used in the text area extraction,and the text is detected as the target.The whole algorithm is built by the neural network.The accuracy of the extraction of the text area is greatly improved,and the accuracy of the position of the entire text is also very high.After getting a large number of text areas,you need to filter the text area to get a useful text area.(3)This article is mainly to obtain basic citizen information on the ID card,such as name,ethnicity,address,ID number,etc.Based on the convolutional network and the cyclic network,the CTC algorithm is introduced,which can realize the nonsegmented end-to-end character recognition of uncertain length text characters.The recognized objects include Chinese characters,numbers and English letters.In view of the above research,the prototype of the ID card text information extraction system based on neural network is realized,which mainly includes the preprocessing module of ID card image,the extraction and screening module of text area and the character recognition module of text.Finally,the results on the dataset show that the system achieves higher accuracy,lower average edit distance and faster recognition speed in the task of extracting ID text information.