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

Imprinted character recognition method and system based on deep learning

A recognition method and character technology, applied in the field of image recognition, can solve problems such as poor robustness and reduced generalization ability, and achieve high accuracy

Pending Publication Date: 2022-07-15
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The traditional method of character recognition uses template matching to locate characters and then segment and recognize them. This method has poor robustness, and its generalization ability drops rapidly when fonts change, blur or background interference.

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
  • Imprinted character recognition method and system based on deep learning
  • Imprinted character recognition method and system based on deep learning
  • Imprinted character recognition method and system based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0031] Embodiment 1 is an image character sequence recognition method, and the method includes the following steps: S1: preprocessing an original image to obtain pixel-level image data to be processed; S2: inputting the pixel-level image data to be processed into a convolutional neural network Perform feature extraction processing to obtain a feature map with a size of N*M*K, that is, a feature map of K channels with a size of N*M; S3: Perform a matrix transposition operation on the N*M*K feature map to obtain M*N *K feature map; tile the M*N*K feature image and convert it into a two-dimensional feature map of size M*(N*K); S4: The feature map of M*(N*K) is used The classification algorithm carries out the classification and discrimination of 1 categories, and obtains the sequence containing M classification results; in the M classification results, each discrimination result is a class in the 1 categories; S5: use the CTC algorithm to the sequence of the M classification resul...

Embodiment 2

[0042] In this embodiment, when the character sequence in the original image is an embossed digital character, a recognition model is obtained by training using the image character sequence recognition method, and the digital character is embossed. Its system block diagram refer to Figure 1 As shown, the system includes a training set acquisition module, a model training module, a to-be-recognized image acquisition module and an identification module.

[0043] (1) Training data acquisition module: use the camera to collect the imprinted character images, and try to avoid tilting when taking pictures. Scale the image to a fixed size (such as 72*288), and label the characters on the image to get the original dataset. The original dataset is divided into three datasets: training set, validation set and test set, which are used for model training, validation and testing respectively. In order to expand the dataset and increase the robustness of the model, it is necessary to per...

Embodiment 3

[0051] This embodiment is the process of using the image character sequence recognition method to train to obtain the recognition model:

[0052] 1. Training data acquisition module: use the camera to collect the imprinted character images, and try to avoid tilting when taking pictures.

[0053] 2. Scale the image to a fixed size (eg 72*288).

[0054] 3. Obtain 30,000 images through the above operations, and select 10,000 images as the test set. Annotate the remaining 20,000 images. First, encode according to the range of the font to be recognized, and then mark the characters on the image according to the encoding of each word. In particular, the code indicating blank should be added to the font code, but it is not necessary to mark when marking.

[0055] 4. Randomly shuffle and split into training set, validation set and test set. There are 15,000 images in the training set and 5,000 images in the validation set.

[0056] 5. Sharpen, blur and add noise to the training s...

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 a method for identifying an image character sequence, which comprises the following steps of: preprocessing an original image to obtain pixel-level image data to be processed; pixel-level image data to be processed are input into the convolutional neural network for feature extraction processing, so that a feature map is obtained, and matrix transposition operation processing is performed on the obtained feature map; classifying and distinguishing the transposed feature map by using a classification algorithm to obtain a sequence containing a plurality of classification results; and processing the sequence of the plurality of classification results by using a CTC algorithm to obtain a sequence with J characters. According to the method, the existing convolutional neural network is improved, and the efficient, accurate and stable image character sequence recognition, especially the recognition of the embossed characters, is realized according to the characteristics of the actually obtained feature map and by combining the neural network, the softmax classifier and the CTC algorithm. The innovation point is that multi-style font recognition is realized through a few-sample and multi-style deep learning training technology.

Description

technical field [0001] The invention relates to the field of image recognition, in particular to an image character sequence recognition method. Background technique [0002] Compared with the background as the character carrier, the embossed characters are convex or concave, often with a certain height or depth, and belong to three-dimensional characters; the embossed characters are easy to wear, and after wear, the characters and the background are of the same material, so There is no obvious color difference from the background, and it belongs to the character without color difference; the font size and style of the characters pressed by different fonts may be different; even if the same font is pressed, the same character is formed due to the influence of on-site processing factors such as loosening or sign material. The height and shape are also different. These all bring difficulties to the recognition of embossed characters. [0003] Traditional character recognitio...

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): G06V10/764G06V10/82G06V30/10G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/047G06N3/045G06F18/2415
Inventor 张岩张新宇周远恺
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA
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