Bill classification and bill field extraction method based on deep learning and OCR

Abstract

A deep learning and OCR-based bill classification and bill field extraction method disclosed by the present invention comprises the following steps of selecting a plurality of standard bill images, intercepting the external rectangular contours of the office seal images of the standard bill images as the training samples of the deep learning, and obtaining a deep learning model; classifying the external rectangular contours of the office seals of the to-be-identified bill images via the deep learning model, if the classification is successful, outputting the bill types of the to-be-identifiedbill images and determining the to-be-identified bill images as the standard bill images, if the classification is unsuccessful, determining the to-be-identified bill images as the non-standard bill images. The method of the present invention can classify the bills automatically and extract the bill content intelligently, provides the services of retrieving rapidly and obtaining the bill information rapidly to the financial staff, and enables the working efficiency to be improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a bill classification and bill field extraction method based on deep learning and OCR. Background technique [0002] OCR-based bill information extraction technology refers to the conversion of unstructured bill images into structured data with the help of OCR technology to realize the extraction of bill information. At present, there are roughly two types of bill OCR processing: (1) manual entry of bill images, which requires manual operations by employees, is time-consuming and laborious, and has high labor costs. Human error; (2) Automatic recognition of bills with simple formats, generally only simple and fixed-format bills can be recognized; [0003] Chinese Patent Publication No. CN204965587 discloses a financial bill recognition and classification storage device, which realizes the front and back image recognition device of financial bills through an internal turning wheel...

AI Technical Summary

Problems solved by technology

At present, there are roughly two types of bill OCR processing: (1) manual entry of bill images, which requires manual operations by employees, is time-consuming and laborious, and has high labor costs. Human error; (2) Automatic recognition of bills with simple formats, generally only simple and fixed-format bills can be recognized;

[0003] Chinese Patent Publication No. CN204965587 discloses a financial bill recognition and classification storage device, which realizes the front and back image recognition device of financial bills through an internal turning wheel, but this device requires manual classification and can only be used for financial and bank bills. Applicable bill types are not wide enough

However, the system cannot adaptively distinguish the fields of set-printing and machine-printing in the bill, and uses different methods to realize field positioning, resulting in low recognition accuracy

Images