Research On The Extraction Method Of Important Information Of Express Bill Image Based On OCR

In recent years,with the advent of e-commerce,China’s express delivery industry has continued to develop rapidly,online purchase has become the mainstream purchase method,and express delivery orders have gradually changed from traditional handwritten express orders to electronic express orders.At present,the internal resource sharing and interconnection of logistics industry companies are not perfect enough,and some ends still enter express data through manual records,which affects work efficiency to a certain extent,and labor costs are very high.By intelligently identifying the important information data of the express order,it can not only improve the terminal efficiency and reduce operating costs,but also realize the unified management of the important information of the express bill and the development of the express delivery industry.The OCR-based express bill image important information extraction method can significantly improve the work efficiency of logistics related practitioners,and the algorithm proposed in this paper has the advantages of high text recognition accuracy and can improve the problem of ambiguity of express bill in some realities.Through multiple experimental comparisons,the reliability and effectiveness of the method of extracting important information from the express bill image in this paper are fully illustrated,which has broad application prospects and application value in the express delivery industry,and has good practical significance.The work of this paper mainly includes:Firstly,the purpose and significance of this research project are summarized,the difficulties of the current express bill identification are analyzed,the key technologies of important information extraction of express bill are expounded,several super-resolution reconstruction methods are compared,the advantages and disadvantages of each method are analyzed,and finally the text super-resolution network(TSRN)is selected to improve.For the text detection part,the current object detection algorithm with better performance is studied,the characteristics of text detection are analyzed,and the CTPN(Connection Text Proposal Network)algorithm that is more suitable for text detection is selected.The text recognition method based on attention mechanism and the text recognition method based on connection time series classification(CTC)are compared and analyzed,and the text recognition algorithm that is more suitable for this paper is selected for the text characteristics on the express bill image studied in this paper.Secondly,the problem of image blurring of express bill is studied,the super-resolution reconstruction algorithm is used to improve the resolution of express single image,the network structure of TSRN is studied,and information distillation block(IDB)is added on the basis of TSRN network to enhance shallow feature information,so as to improve the TSRN network.The effectiveness of the proposed algorithm is proved by experimental comparison of the improved algorithm on the dataset.Thirdly,the text detection algorithm based on improved TSRN and CTPN is studied,and the image output of the improved TSRN algorithm is used as the input of the CTPN text detection algorithm,and the network structure of the CTPN algorithm is studied.The effect of the algorithm is verified by selecting some images,and it is proved that the effect of image text detection is better after the improved super-resolution reconstruction algorithm.Finally,the text recognition algorithm based on improved TSRN is studied,and the output image after the improved super-resolution reconstruction algorithm is used as the input of the text recognition part.In the text recognition part,the network structure of the text recognition algorithm is analyzed,and the effect of the algorithm is verified by selecting some express bill images,which proves that after improving the super-resolution reconstruction algorithm,the recognition accuracy of important information of express order images can be improved,which also improves the work efficiency of express delivery staff.And the algorithm can be widely used in fields such as invoice or ID card recognition,with good robustness and generalization.