OCR identification method based on model fusion

A recognition method and model fusion technology, applied in character and pattern recognition, neural learning methods, biological neural network models, etc., can solve the problems of low recognition accuracy and positioning deviation, and achieve complementary advantages, improve accuracy, and improve Effects of working with sequence data

Inactive Publication Date: 2020-09-01
CENT SOUTH UNIV
View PDF8 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In the traditional OCR recognition method, image processing technology is usually used to locate the area to be recognized one by one, but when the picture is more complex, the positioning will have a large deviation, so the accuracy of subsequent recognition is very low.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • OCR identification method based on model fusion
  • OCR identification method based on model fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] This example provides an OCR recognition method based on model fusion. This method uses several recognition models to build a deep learning neural network model. The recognition models train the character positioning data respectively, and obtain the pre-output recognition results after the model converges. The above pre-output recognition results are fused and optimized to output the final recognition results.

[0041] The method flow is as follows figure 1 As shown, it specifically includes the following steps:

[0042] (1) Obtain the picture data to be recognized, and preprocess the picture data.

[0043] (1.1) Preprocess the picture: correct the tilt, grayscale, sharpen, adjust the contrast and brightness of the picture, and scale the picture to a uniform size.

[0044] (1.2) Normalize the picture. m=M / 255. Among them, M is the original data, and m is the result after normalization. After normalization, the image pixel value range is [0, 1], which is more conduc...

Embodiment 2

[0062] This example uses the OCR recognition method based on model fusion in Embodiment 1 to perform recognition processing on a picture to be recognized.

[0063] The complete process steps are as follows:

[0064] (1) Input a picture to be recognized and preprocess it, including correcting the skewed and distorted picture, and removing noise from the blurred picture to make it clearer.

[0065] (2) Use the pixel-link neural network to locate the area containing characters in the picture, take a screenshot and save it.

[0066] (3) Perform random data enhancement on the character picture to be recognized, and randomly enhance and expand from one original picture to be recognized to 30 pictures. Specific enhancement methods include: random up and down translation, left and right translation, random rotation at a certain angle, random addition of salt and pepper noise, random adjustment of luminosity, brightness, contrast, etc.

[0067] (4) Randomly assign 10 pictures to each...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention provides an OCR identification method based on model fusion. According to the method, a plurality of recognition models are adopted to build a deep learning neural network model, the recognition models train character positioning data respectively until the models converge to obtain a pre-output recognition result, and the pre-output recognition result is fused and optimized to output a final recognition result. The multi-model fusion OCR recognition problem is solved, and compared with the traditional OCR recognition method, the deep learning multi-model fusion OCR recognition technology is adopted, so that positioning is accurate, the recognition accuracy is high, and the model recognition effect is more stable.

Description

technical field [0001] The invention belongs to the technical field of image and word processing, and in particular relates to an OCR recognition method based on model fusion. Background technique [0002] With the rapid development of social economy, people generate a large amount of data in the form of pictures every day, including scanned ID cards, automatically taken license plate photos, electronic invoices, scanned documents, etc., and the character information in these pictures is automatically The process of converting it into text form is called Optical Character Recognition (OCR). In the traditional OCR recognition process, the general process is manual data collection, manual data entry, and manual verification. However, the entire process not only takes a long labor cycle, but also is prone to manual data entry errors. In view of this situation, an efficient and fast OCR recognition method is urgently needed. [0003] OCR has been applied to various fields, suc...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/20G06K9/34G06N3/04G06N3/08
CPCG06N3/08G06V10/22G06V30/153G06N3/045
Inventor 张昊傅枧根
Owner CENT SOUTH UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap