Document identification method and identification system

Abstract

The embodiment of the invention relates to a receipt identification method and identification system. According to the document identification method, firstly, image enhancement processing is carried out on a document picture; then carrying out quality scoring on the enhanced receipt pictures by adopting a twin neural network, and determining the receipt pictures of which the quality scores are not lower than a preset threshold value as to-be-identified receipt pictures; a first CNN being used to position four corner points of the to-be-identified receipt picture, and a to-be-identified area being extracted along the four corner points; performing textbox detection on the to-be-identified area by adopting FPN to obtain a plurality of textboxes; performing text recognition on the plurality of text boxes obtained by detection by using CRNN and CTC to obtain a character field corresponding to each text box; and finally, identifying the character fields obtained by identification by using TextCNN to obtain character identification results corresponding to the character fields. According to the invention, the document identification efficiency and the entry efficiency can be improved, and the entry cost is reduced.

Description

technical field [0001] The embodiments of the present invention relate to the field of image recognition, and in particular to a bill recognition method and a recognition system. Background technique [0002] At present, the Optical Character Recognition (OCR) algorithm has gradually reached a mature stage, and OCR recognition technology is gradually landing in various scenarios, such as ID card recognition, bank card recognition, driver's license recognition, etc. The demand for document identification is also gradually emerging in the market, and it is becoming more and more a future trend to use automatic identification of documents to replace manual input. [0003] There are already some solutions for document recognition, but these methods have exposed some disadvantages in use. Some existing solutions use traditional methods such as text detection and recognition, but the traditional methods have poor generalization ability for lighting and even image quality. , some ...

AI Technical Summary

Problems solved by technology

This move has greatly affected the variety of document formats

In the case of some newly added formats, it will fail; in addition, basically all existing solutions

Images