A Text Detection Post-processing Method Based on Geometric Features

Abstract

The present invention provides a text detection post-processing method based on geometric features. The post-processing method is used to post-process the original prediction frame. The original prediction frame is obtained by a deep learning text detection method. The post-processing method includes the following Steps: S1. Based on the background removal algorithm, remove the redundant background in the prediction frame without changing the intersection area of ​​the prediction frame and the character area; S2. Based on the candidate frame expansion algorithm, expand the obtained prediction frame in a certain order , so that it can completely extract the character area; S3, based on the non-standard frame removal algorithm, remove the non-standard prediction frame; S4, based on the repeated frame removal algorithm, remove the obtained repeated prediction frame to obtain the final prediction frame. This method can effectively solve the problems of character misrecognition, detection deviation, overlapping detection, etc. during text detection, and make the detection results more excellent.

Description

technical field [0001] The invention mainly relates to the related technical field of document image processing, in particular to a post-processing method for text detection based on geometric features. Background technique [0002] Document images are important carriers of information and play an important role in daily life. With the wide application of digitization in various fields, humans hope that machines can also imitate the ability of humans to read books, so the optical character recognition technology OCR (Optical Character Recognition) came into being. And text detection is an indispensable part of character recognition technology OCR, which is very important for subsequent text recognition. Efficient and accurate text detection has important applications in the field of document images, including character recognition systems, multilingual translation of images, and human-computer interaction. [0003] At present, the deep learning text detection method mainly...

AI Technical Summary

Problems solved by technology

However, in practical applications, the final prediction frame is often not very accurate, so many algorithms also perform other post-processing, such as merging adjacent candidate frames

[0004] The current deep learning text detection method has less post-processing, and the obtained prediction frame can generally obtain better detection results, but there are also many details that have not been processed in place.

Images