CRNN-based bill information recognition algorithm, device and storage medium

Abstract

The invention provides a CRNN-based bill information recognition algorithm, a device and a storage medium. The method comprises the following steps of: locating and segmenting a specific area, such asa buyer name area, from a bill picture to facilitate better identification; CRNN neural network is used to train the recognition model, so as to recognize the information of the specific area on thebill. The method is preferably applicable to an algorithm for identifying invoice information. The recognition algorithm of the invention does not need to segment the Chinese character, does not needto set the character length, and does not need to recognize a single Chinese character, thus avoiding the influence on the recognition rate caused by the segmentation error, thereby greatly improvingthe recognition accuracy rate.

Description

technical field [0001] The application belongs to the field of bill picture processing and recognition in the financial field, and specifically relates to the recognition of characters in bill pictures, and in particular to a bill information recognition algorithm, device and storage medium based on a convolutional cyclic neural network (CRNN). Background technique [0002] In recent years, e-commerce information has penetrated into various industries. The scale and business volume of many enterprises have continued to grow, and the capital transactions of enterprises have diversified. The number of financial invoices and other financial instruments has increased sharply, which makes the management and workload of the company's financial system more onerous. On the one hand, financial bill management review, specifically invoice information review and reimbursement, has an increasing demand for electronic and digitalization. In fact, most of the current bill audits are manua...

AI Technical Summary

Problems solved by technology

The traditional recognition method based on a simple CNN neural network can only train and recognize images with a fixed label length, and the length of the text in the bill may not be the same, such as the name of the buyer in the invoice, so a single Chinese character must be segmented for recognition , so that the recognition rate will be affected due to the error in segmentation

In addition, the direct use of RNN neural network to identify the effect is low

[0004] Therefore, the problem with the existing technology is that the commonly used OCR system can only complete the scanning and entry of some types of invoices, and the recognition accuracy is not high; while the detection and recognition of text in different scenarios, the recognition of text information such as Chinese characters, characters and numbers , The end-to-end solution technical solution also needs to cut the text, the steps are complicated and easy to cause error accumulation, low efficiency and many other shortcomings, which need to be improved

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the inventors of the present invention expect to provide an algorithm and equipment based on a more optimized CRNN neural network model to quickly and accurately identify text information in bills, especially a specific area in bills The text information recognition algorithm overcomes the above-mentioned deficiencies in the existing technology, and frees financial personnel from the tedious bill review work

Images