Research On Bill Number Information Recognition Algorithm And Its System Implementation

Domestic bills are important accounting bookkeeping and tax payment documents of enterprises,which have strict unified format.Many enterprises' financial centers need to process a large number of bill information every day,but the manual processing efficiency is low.In order to facilitate the extraction,storage and transmission of invoice information,it is urgent to need an automated bill processing system to reduce the processing cost and improve the financial operation ability.Bill information processing is the process of image acquisition,image preprocessing,character segmentation and character recognition.The effective information on the invoice can be extracted in time,which can effectively solve the problem of huge workload and low efficiency caused by manual processing.In this paper,the bill image is taken as the research object.Through the research of information extraction algorithm in complex environment,the automatic recognition of bill information is realized.The details are as follows:(1)The types of light sources and their application are studied.Different requirements and target characteristics are given.Light sources and lighting methods are selected.The hardware system of bill imaging is built;(2)Aiming at the problem that the overlapping of chapter code can not be recognized,the relationship between light and imaging medium color is studied,and the monochromatic light back transmission lighting scheme is proposed;(3)The prior information is used to locate the target area quickly,a progressive number location method based on bidirectional projection is proposed to realize the accurate segmentation of single character;(4)Based on the transfer learning method of handwritten characters,the deep neural network learning model of bill number is established,and the accurate recognition of invoice number under small sample is realized.On the basis of the above research,the bill number automatic recognition system developed in this paper,through the experimental verification,its recognition accuracy can reach 94.9%,which improves the level of invoice automatic processing.