Unlock instant, AI-driven research and patent intelligence for your innovation.

Character recognition training system and method

A text recognition and training system technology, applied in the field of text recognition, can solve the problems of difficulty in accurate recognition, easy confusion, different shapes and meanings of Chinese characters, etc., to achieve the effect of improving work efficiency

Inactive Publication Date: 2019-05-14
CHUSUDU (SUZHOU) TECH CO LTD
View PDF13 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, Chinese text recognition is significantly different from English text recognition in the complexity of the task. First, there is a huge difference in the number of characters. English only needs to recognize 26 letters, but Chinese only has three or four thousand commonly used fonts; and , many Chinese characters are similar in shape but have very different meanings, which also brings difficulties to accurate recognition.

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 training system and method
  • Character recognition training system and method
  • Character recognition training system and method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] Character recognition training system of the present invention such as image 3 As shown, it includes a preprocessing unit, a feature extraction unit, a text recognition unit, a loss function, and a mapping unit; the feature extraction unit is specifically a convolutional neural network CNN, and the text recognition unit is specifically a cyclic neural network RNN.

[0035] The preprocessing unit needs to (1) label the training sample set, that is, pictures including text content, where the labeling specifically refers to the identification of specific text; (2) the labeling of each picture in the training set on the text category in the text library, The text category contained in the picture is not marked as 0, and the text category not contained in the picture is marked as 0. details as follows:

[0036] If the picture contains the text "the road ahead goes straight", then each Chinese character of "front", "square", "road", "road", "straight" and "row" corresponds ...

Embodiment 2

[0058] Character recognition training system of the present invention such as Figure 4 As shown, it includes a preprocessing unit, a feature extraction unit, a text recognition unit 1, a loss function 1, a text recognition unit 2, a loss function 2, and a mapping unit; the feature extraction unit is specifically a convolutional neural network CNN, and the text recognition unit is specifically Recurrent Neural Network RNN.

[0059] The preprocessing unit needs to (1) label the training sample set, that is, pictures including text content. The labeling here specifically refers to identifying the specific text and ensuring the order; (2) each picture in the training set corresponds to the category in the text library If the text category is included in the picture, the label is not 0, and the text category that is not included in the picture is marked as 0. The specific classification is as follows:

[0060] If the picture contains the text "the road ahead goes straight", then ...

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 training system and method, and belongs to the character recognition technology. Existing technology, only a sequence loss function is used in a neuralnetwork training process. According to the character recognition system and the character recognition method, the loss function adopts the sequence loss function and the classification loss function,and classification errors in the Chinese character recognition process are effectively solved.

Description

technical field [0001] The invention relates to a character recognition technology, in particular to a Chinese character recognition training method. Background technique [0002] Most of the current text recognition training structures based on deep learning are as follows: figure 1 As shown, image features are first extracted by a feature extraction model such as a convolutional neural network, and then a text sequence result is generated using a recurrent neural network or natural language processing method, and the loss function of the sequence model is used for alignment and loss calculation. During the training process, the feature extraction model is indirectly adjusted through the sequence loss function so that it can extract the most expressive features. This has yielded very good results in text recognition models for English. However, Chinese text recognition is significantly different from English text recognition in the complexity of the task. First, there is ...

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): G06K9/34G06K9/62G06N3/04
Inventor 胡杰
Owner CHUSUDU (SUZHOU) TECH CO LTD