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

Character recognition method and system based on attention mechanism

A text recognition, attention technology, applied in character recognition, character and pattern recognition, computer parts and other directions, can solve the problem of forming noise area, limited attention area, attention drift and so on

Pending Publication Date: 2020-10-16
厦门商集网络科技有限责任公司 +1
View PDF0 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] However, the existing deep learning model based on the attention mechanism has two defects: (1) Due to the limited focus area of ​​the attention on the feature map, the area that has not been paid attention to during the training phase will form a noise area in the feature map
The attention generated by the attention module is easily disturbed by the noise area, and cannot be well focused on the area where the text is located, resulting in wrong text recognition, that is, the problem of "attention drift"; (2) the text to be recognized often has a strong contextual relevance
In the existing technology, only forward recognition and decoding are used, and the model can only decode from front to back, so that the first decoded characters often lack context information. When the characters are difficult to recognize, the first decoded characters are prone to errors

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
  • Character recognition method and system based on attention mechanism
  • Character recognition method and system based on attention mechanism
  • Character recognition method and system based on attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0070] like figure 1 As shown, a text recognition method based on attention mechanism includes the following steps:

[0071] S1: Build a character recognition model for recognizing characters in images; the character recognition model consists of the following modules:

[0072] Convolutional neural network for extracting feature maps of input images;

[0073] An attention mechanism module, including a sequence encoder, a forward sequence decoder and a reverse sequence decoder, is used to encode and decode the feature map, and output the feature vector of the predicted character;

[0074] The character decoding layer is used to compile the feature vector of the predicted character into a text recognition result, and at the same time compile the feature map into a feature map character probability vector;

[0075] S2: build a training sample set, the training sample set includes a training image and an image label corresponding to the training image, wherein the image label is...

Embodiment 2

[0084] like figure 1 As shown, a text recognition method based on attention mechanism includes the following steps:

[0085] S1: Construct a character recognition model for recognizing characters in images; the character recognition model consists of a convolutional neural network, an attention mechanism module, and a character decoding layer, wherein the attention mechanism module includes a sequence encoder, a regular Towards sequence decoder and reverse sequence decoder.

[0086] In the S1 step, the convolutional neural network includes a multi-layer convolutional filter bank and a pooling sub-module, the convolutional filter bank adopts a residual structure, and the character decoding layer is fully connected by a multi-layer neural network layer structure, wherein the multi-layer convolutional filter bank extracts image features, the pooling sub-module changes the resolution of the feature map, and the output of the convolutional neural network is a feature map with a ce...

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 relates to a character recognition method and system based on an attention mechanism, and relates to a deep learning and image processing technology. According to the method, a convolutional neural network and a linguistic module based on an attention mechanism are used as the backbone of a deep learning model, a customized loss function is used to reinforce feature map extraction, the model is guided to learn to distinguish a foreground and a background during training, and forward and reverse bidirectional decoders are introduced to perform bidirectional decoding on characters.The method and the system are high in anti-interference capability, attention drifting can be reduced, and meanwhile, the situation that final recognition fails due to the fact that the first character is difficult to recognize during the forward decoding of the model can be avoided.

Description

technical field [0001] The present invention relates to deep learning and image processing technology, in particular to a text recognition method and system based on an attention mechanism. Background technique [0002] There are many existing text recognition technologies, including traditional OCR recognition methods and methods based on deep learning. The method based on deep learning inputs a large number of manually labeled image and text samples into the designed neural network, so that the parameters in the neural network can be trained to fit the mapping relationship between the image and the text, and then complete the recognition task. The methods of deep learning are mainly divided into methods based on attention mechanism and methods based on CTC. Among them, the attention mechanism in deep learning (https: / / blog.csdn.net / hpul fc / article / details / 80448570) is essentially similar to the selective visual attention mechanism of human beings. Select the information ...

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
IPC IPC(8): G06K9/00G06K9/46G06K9/62G06N3/04
CPCG06V30/40G06V10/40G06V30/10G06N3/048G06N3/045G06F18/214
Inventor 顾澄宇王士林陈凯周异何建华
Owner 厦门商集网络科技有限责任公司
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