OCR identification method based on localization platform and identification system thereof

Abstract

The invention discloses an OCR identification method based on localization platform and identification system thereof, and belongs to the technical field of pattern recognition and computer vision. The method comprises the following steps: arranging characters to be recognized, making a data set according to the collected characters, and performing data augmentation processing; establishing a deep learning framework based on an image processing tool of a localization platform, and establishing an OCR network structure; performing image size normalization on the to-be-trained data by using a compiled image processing tool, setting training parameters for training, and performing iterative optimization on the parameters of the network structure in the training process; and finally, sending the processed picture into the trained network structure to calculate a most matched label result, and outputting an OCR identification result. According to the method, confidentiality and autonomous controllability of the technology are achieved, the character recognition effect is good, and high robustness is achieved for bankbook, invoice and deposit receipt type font recognition.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition and computer vision, in particular to an OCR recognition method and a recognition system based on a localized platform. Background technique [0002] In recent years, with the rapid development of computer vision technology, the OCR recognition of pictures and texts has become a popular direction. At present, conventional text recognition, text recognition in natural scenes, ID card recognition in the security field, bank card recognition in the financial field, etc. OCR recognition research There are many, and there are already mature applications. The realization of OCR recognition technology mainly depends on the deep learning framework, such as TensorFlow, caffe and Paddle and other frameworks. The recognition network model is built in the framework, and the model is optimized and trained through a large amount of data, so as to realize the text recognition through the model. At t...

AI Technical Summary

Problems solved by technology

[0003] After long-term practice by the applicant, it is found that there are the following deficiencies in the existing technology: 1. The OCR recognition research on the domestic platform cannot directly use the framework of foreign platforms for OCR recognition research due to the different architectures of different platforms

However, there are problems with the confidentiality and controllability of technology and data information when directly using foreign platforms

2. Units using localized equipment need to manually check and extract text when they need to perform text verification and information extraction, which not only has high labor costs but also low efficiency

3. At present, the existing OCR recognition system is not compatible with different application scenarios, and has high requirements for image quality. At the same time, clear and coherent printing fonts are also required.

In some application scenarios, such as invoices, deposit slips, etc., the dot matrix printers used for text printing are different from ordinary laser printers, which affects the recognition accuracy.

Images