Text recognition method based on optical character recognition and error correction tight coupling processing

Abstract

The invention relates to a text recognition method based on optical character recognition and error correction tight coupling processing. According to the method, a text image is recognized through aneural network model, an optimal candidate text sentence is selected through text bundle search transcription and lexicon selection processing by utilizing internal information generated during text image recognition, error correction is performed through the neural network model, and a more accurate text recognition result is output. Optical character recognition and text error correction are tightly coupled, and compared with an existing loose coupling text recognition method, the text error correction performance can be effectively improved, and the text recognition accuracy is improved.

Description

technical field [0001] The invention relates to a text recognition method based on optical character recognition and error correction tightly coupled processing, belonging to the field of OCR recognition. Background technique [0002] With the information processing technology in recent years, the performance of the optical character recognition (OCR) system based on machine deep learning for text positioning and text recognition has been greatly improved. In some fields, the accuracy of text recognition is close to the level of manual recognition, which helps Realize the application of various scenarios, such as ID card recognition and license plate recognition. In some commercial applications, such as bill reimbursement and bank transactions, OCR technology is also playing an important role. OCR recognition requires error correction for the recognition results to ensure the correctness of the results. Using machine automatic text error correction is an important way. At p...

AI Technical Summary

Problems solved by technology

[0003] In the existing technology, the two links of OCR recognition and error correction processing based on machine deep learning are loosely coupled, such as figure 2 As shown, the OCR recognition module outputs a string of text, and the text error correction module uses the text string as input to correct possible recognition errors. Except that the text error correction module takes the output of the OCR recognition module as input, the difference between the two modules There is no other correlation between them. This loosely coupled relationship makes the recognition accuracy not high, and it is difficult to be competent for complex recognition scenarios.

Images