Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Natural scene multilingual text detection method based on deep learning

A natural scene and text detection technology, applied in the field of image processing, can solve problems such as difficult to deal with blurred scene text, multi-processing steps, difficult to distinguish, etc., and achieve the effect of convenient recognition and processing and good robustness

Active Publication Date: 2017-03-15
NANJING UNIV
View PDF3 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in practical applications, connected regions in multiple languages ​​are difficult to distinguish from cluttered backgrounds
The latter includes SWT and sliding window methods, the SWT method is more difficult to deal with blurred scene text, and the sliding window requires more processing steps for skewed text

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
  • Natural scene multilingual text detection method based on deep learning
  • Natural scene multilingual text detection method based on deep learning
  • Natural scene multilingual text detection method based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0037] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific preferred embodiments.

[0038] Such as figure 1 As shown, a deep learning-based method for multilingual text detection in natural scenes includes the following steps.

[0039] Step 1, convert to a grayscale image: input the image to be detected, and convert the input image to be detected into a grayscale image I.

[0040] Step 2, construct the ER tree: obtain several binary images from the grayscale image I converted in step 1 according to the threshold value, each binary image corresponds to several ERs, pass all ERs through their inclusion relationship in the binary image , and form it into a tree, which is called ER tree; the ER tree satisfies the following properties:

[0041]

[0042] Among them, Ri and Rj are two connected nodes on the ER tree, and Ri represents the parent node, and Rj represents the child node; both p and q represent th...

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 discloses a natural scene multilingual text detection method based on deep learning, which comprises the following steps: converting a text into a gray image, constructing an ER tree, obtaining LERs, removing repeated LERs, distinguishing the text and the background impurity, performing seed growth and clustering, and identifying the text with a random forest, wherein, a two-stage CNN is used in the distinction between characters and background impurities, wherein a first-stage CNN is used for distinguishing between the background impurity and character-similar symbols Symbol, a second-stage CNN is used for dividing the character-similar symbols Symbol into Chinese, English and identifies Sign, and removing the identifies Sign so that the components containing Chinese and English are preserved. After the method is used, it is possible to detect the multilingual text and the broken characters in a natural scene. In addition, compared with the prior art, the method has good robustness in the natural scene, and can be applied well to the multilingual complex environment, and is not lower than the detection speed in the prior art.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for detecting multilingual text in natural scenes based on deep learning. Background technique [0002] At present, multilingual text detection methods in natural scenes are mainly divided into MSER methods and non-MSER methods. The former mainly uses MSER to extract connected regions in multilingual texts. However, in practical applications, it is difficult to distinguish connected regions in multiple languages ​​from cluttered backgrounds. The latter include SWT and sliding window methods, the SWT method is more difficult to deal with blurred scene text, and the sliding window requires more processing steps for skewed text. A new type of text extraction method - Linked Extremal Region (LER) is used in this technology. This method can extract multilingual text in the form of non-connected regions, which is convenient for subsequent recognition processing. T...

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): G06K9/34
CPCG06V10/267G06V10/273
Inventor 路通刘若泽
Owner NANJING UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com