Method for automatically correcting error of character recognition result

Abstract

The invention discloses a method for carrying out automatic error correction on a character recognition result, which comprises the following steps of: acquiring predicted character information in an image to be processed and standard character information in the same scene; searching candidate common subsequences in the predicted character information and the standard character information; traversing the candidate common sub-sequence, judging whether the candidate common sub-sequence is a reasonable sub-column or not, and if yes, ending traversing; filtering and filling the predicted character information by using the common subsequence to obtain processed predicted character information; and correcting the predicted character information according to the processed predicted character information, and outputting correct character information. The method is suitable for post-processing an OCR detection recognition result, and can effectively solve the problem of result errors caused by character false detection, character missing detection and character false recognition.

Description

technical field [0001] The invention belongs to the technical field of optical character recognition, in particular to an automatic error correction method for character recognition results. Background technique [0002] For production scenarios such as video packaging, pharmaceutical packaging, and canned or boxed beverages that are common in daily life, the products produced on the same day all have the same character inkjet content, but during the production process, due to the jitter of the production line speed, the inkjet The occasional situation of the coding machine leads to problems with the coding of the products produced at a certain time. For example: the code cannot be recognized, the content of the code is wrong, and the code is missed. [0003] The solution to this type of character detection and recognition results, whether it is traditional image processing or the use of convolutional neural networks to detect and recognize characters, will face a problem, ...

AI Technical Summary

Problems solved by technology

[0002] For production scenarios such as video packaging, pharmaceutical packaging, and canned or boxed beverages that are common in daily life, the products produced on the same day all have the same character inkjet content, but during the production process, due to the jitter of the production line speed, the inkjet The occasional situation of the coding machine leads to problems with the coding of the products produced at a certain time

For example: the code cannot be recognized, the content of the code is wrong, the code is missing, etc.

[0003] The solution to this type of character detection and recognition results, whether it is traditional image processing or the use of convolutional neural networks to detect and recognize characters, will face a problem, that is, there are false detections or missing detections for normal products. When the imaging is unstable due to changes in external factors such as the production environment and lighting, the probability of false detection or missed detection will be greater

Using the calculation of the IOU value between the detection frames, and then filtering the unreasonable detection results by setting the threshold can solve the problem that occurs when the detection frames overlap, but it is powerless for the discrete frames that are detected incorrectly, and for the characters that are missed and misrecognized. can't handle

In this case, a considerable number of normal products will be rejected as abnormal products, which will greatly increase production costs

Images