CNN-based handwritten Chinese text recognition method

A text recognition and Chinese technology, applied in character and pattern recognition, instruments, computer parts, etc., can solve the problems of low recognition accuracy and achieve high accuracy

Pending Publication Date: 2019-05-10
TIANJIN UNIV
View PDF2 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The present invention provides a handwritten Chinese text recognition method based on CNN. The present invention solves the problem of low accuracy of traditional handwritten Chinese recognition, a

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
  • CNN-based handwritten Chinese text recognition method
  • CNN-based handwritten Chinese text recognition method
  • CNN-based handwritten Chinese text recognition method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] The embodiment of the present invention provides a CNN-based handwritten Chinese text recognition method, see figure 1 , the method includes the following steps:

[0041] 101: Grayscale and binarize text images, and then use histogram projection [5,6] Segment Chinese text; first segment a single line of text by horizontal scanning, and then segment a single text by vertical scanning;

[0042] 102: Scan and process a single Chinese image, perform orthorectification on the Chinese, and place it in the middle of the image, leaving 10 blank pixels at the top, bottom, left, and right;

[0043] 103: Construct a convolutional neural network, use the training set for training; input the picture to be tested, and use the constructed convolutional neural network for recognition.

[0044] Wherein, the embodiment of the present invention is based on the TensorFlow framework [7] (This framework is known to those skilled in the art) Construct a convolutional neural network, this c...

Embodiment 2

[0061] The scheme in embodiment 1 is further introduced below in conjunction with specific calculation formulas and examples, see the following description for details:

[0062] 201: Binarize the image, project the binarized image in the horizontal direction, calculate pixels, and obtain a projection curve in the horizontal direction;

[0063] Among them, some zero values ​​in the projection curve represent blank areas between text lines, and the text image is segmented with the zero-value central horizontal line to extract each line of text.

[0064] 202: Using the projection curve in the horizontal direction obtained in step 201 to cut the binarized image into rows, and then projecting each row of images in the vertical direction to obtain a projection curve in the vertical direction;

[0065] During the specific implementation, since there may be continuous strokes between the handwritten characters, the projection curve in the vertical direction is often not zero between t...

Embodiment 3

[0105] Combine below Figure 2-Figure 7 , Table 1 and Table 2 carry out feasibility verification to the scheme in embodiment 1 and 2, see the following description for details:

[0106] 301: The embodiment of the present invention uses the softmax cross-entropy function to calculate the loss function, compares the output value of the previous layer with the input real label, calculates the loss and normalizes it to a probability value, and then calculates the mean value; uses the argmax function to find The predicted label is compared with the input real label to calculate the average prediction accuracy rate; dynamically adjust the learning rate, the initial learning rate is set to 0.0002, the number of steps per iteration is 2000, and the attenuation ratio is 0.97. The adamOptimizer optimization algorithm is used to facilitate the dynamic adjustment of parameters.

[0107] 302: First use the constructed convolutional neural network to train and test the HWDB data set, and o...

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 CNN-based handwritten Chinese text recognition method. According to the method, single handwritten Chinese recognition and a character segmentation algorithm are combined, automatic recognition of handwritten Chinese texts is achieved, and the method comprises the following steps that graying and binarization processing are conducted on text pictures, and then histogram projection is used for segmenting the Chinese texts; First, single-row characters are segmented through transverse scanning, and then single characters are segmented through longitudinal scanning. Carrying out scanning processing on a single Chinese picture, carrying out ortho-rectification on the Chinese picture, enabling the Chinese picture to be located in the middle of the picture, and leaving10 blank pixels in the upper, lower, left and right directions respectively; a convolutional neural network comprising four convolutional layers, four pooling layers and two full connection layers isconstructed based on a TensorFlow framework, and a training set is used for training; and inputting a to-be-tested picture, and performing recognition according to the constructed convolutional neuralnetwork.

Description

technical field [0001] The invention relates to the field of computer image processing, in particular to a CNN (convolutional neural network)-based handwritten Chinese text recognition method. Background technique [0002] Handwritten Chinese recognition is one of the hotspots in the field of computer image and vision research. It has been widely used in the recognition of historical documents, mail classification, transcription of handwritten notes, etc. Although domestic and foreign scholars have done a lot of research in this field in the past few decades, there are still many problems that have not been effectively resolved. The main difficulties in handwritten Chinese recognition come from the complex structure of Chinese, many types of characters, large data, different styles of each person, and handwriting distortion. In some cases, the difference between unconstrained handwritten samples between similar Chinese characters can be very small, such as the characters "天...

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): G06K9/34G06K9/62
Inventor 何凯黄婉蓉冯旭高圣楠
Owner TIANJIN 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