Natural scene text detection method based on deep learning

A text detection and natural scene technology, applied in the field of natural scene text detection based on deep learning, to achieve the effect of improving precision and recall rate, suppressing false positive results, and reducing false detection

Active Publication Date: 2020-02-18
SOUTH CHINA UNIV OF TECH
View PDF4 Cites 25 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] In order to perform text detection in natural scenes more accurately and efficiently, and solve the detection problem of arbitrary-shaped text in natural scenes, the present invention proposes a natural scene text detection method based on deep learning

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

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0057] A method for detecting arbitrary-shaped text in natural scenes based on deep learning, such as figure 1 shown, including the following steps:

[0058] S1. Construct and train a neural network-based natural scene text detection model, such as figure 2 As shown, the natural scene text detection model includes a feature extractor, a region proposal network, a RoIAlign pooling layer, and a Mask RCNN head; including the following steps:

[0059] S1.1. Construct a feature extractor based on Feature Pyramid Networks (FPN), and extract the five-level feature maps of C1-C5 from the skeleton network ResNet-50 through the bottom-up path, and through the top-down path and The horizontal connection obtains a total of 5 levels of feature maps from P2-P6, where P6 is obtained from P5 through the maximum pooling with a step size of 2;

[0060] Feature Pyramid Networks (FPN) uses ResNet-50 as the skeleton network; Feature Pyramid Networks (FPN) uses a total of 5 levels from P2 to P6;...

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 an arbitrary-shape natural scene text detection method based on deep learning. The objective of the invention is to improve the accuracy and recall rate of text detection of natural scenes of any shape. The method comprises the following steps: constructing and training a natural scene text detection model based on a neural network, and detecting a natural scene text in a given image by using the trained natural scene text detection model. According to the method, a general integrity perception loss function is provided to replace a regression branch smooth l1 loss function, and an index related to the IoU is directly used as an optimized object, so that the detection precision and recall rate are directly improved; according to the method, the TextIoU head is addedin the Mask RCNN head to predict the IoU of the text mask, the predicted value is multiplied by the classification confidence of the Box head to obtain the final confidence score, and the final detection result is obtained by screening according to the final confidence score. By adding the TextIoU head, a false positive result can be effectively inhibited, so that the detection effect is improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for detecting text in natural scenes based on deep learning. Background technique [0002] Scene text detection is an important branch of target detection and a popular research field of computer vision, which is widely used in image search, automatic driving, blind assistance, real-time translation and other scenarios (Y.Zhu, C.Yao, and X.Bai, "Scene text detection and recognition: Recent advances and future trends," Frontiers of Computer Science, vol.10, no.1, pp.19–36, 2016.). Precise text localization is an important prerequisite for subsequent recognition tasks. However, there are still some difficulties in detecting text in natural scenes. [0003] Diversity and variability of text in natural scenes: Compared with text in documents, text in natural scenes has drastic scale transformations, presenting diversity in fonts, colors, shapes, directions, and la...

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): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V20/00G06N3/045G06F18/24
Inventor 刘发贵谷典
Owner SOUTH CHINA UNIV OF TECH
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