Model generation system and model generation method

Patent metadata

Field	Value
Publication number	US-12380718-B2
Application number	US-202318111254-A
Country	US
Kind code	B2
Filing date	Feb 17, 2023
Priority date	Mar 22, 2022
Publication date	Aug 5, 2025
Grant date	Aug 5, 2025

Abstract

Provided is a model generation system for generating a text line recognition model that recognizes a text line included in a text line image, the model generation system including a processor section, in which the text line recognition model includes a visual feature extractor and a language context relation network, the processor section determines a variable of the language context relation network by acquiring text data for training and thus training the language context relation network by using the acquired text data, determines a variable of the visual feature extractor by training the text line recognition model through the use of a labeled text line image while the variable of the language context relation network is fixed, and generates the text line recognition model while the variable of the language context relation network is set to the determined variable thereof and the variable of the visual feature extractor is set to the determined variable thereof.

First claim

What is claimed is:

1. A model generation system for generating a text line recognition model that recognizes a text line included in a text line image, the model generation system comprising:
a memory; and
a processor section, wherein
the text line recognition model includes
a visual feature extractor that, when executed by the processor section, outputs image feature values from the text line image, and
a language context relation network that, when executed by the processor section, inputs the feature values outputted from the visual feature extractor, and outputs the text line,

the processor section by executing a program stored in the memory that performs the following steps:
(1) determining a variable of the language context relation network by acquiring text data for training and thus training the language context relation network by using the acquired text data,
(2) determining a variable of the visual feature extractor by training the text line recognition model through use of an existing labeled text line image while the variable of the language context relation network is set according to step (1), and
(3) generating the text line recognition model while the variable of the language context relation network is set according to step (1) and the variable of the visual feature extractor is set according to step (2),
wherein the memory is configured to store the text line recognition model.

2. The model generation system according to claim 1 , wherein
the processor section adjusts the variable of the text line recognition model by training the text line recognition model through use of labeled text line images smaller in number than a predetermined number.

3. The model generation system according to claim 1 , wherein
the model generation system is connected to the Internet, and
the processor section accesses the Internet to acquire the text data for the training.

4. The model generation system according to claim 3 , wherein
the text data for the training is formed by copyright-free text data published on the Internet.

5. The model generation system according to claim 2 , wherein
the processor section by executing a program stored in the memory that:
receives a text line image and a label to be attached to the text line image that are inputted by a user, and
adjusts a variable of the text line recognition model by training the text line recognition model through use of the received text line image and label.

6. The model generation system according to claim 1 , wherein
the processor section trains the language context relation network by acquiring text line data for the training, performing word embedding for quantifying the acquired text line data, convolving the quantified data, and inputting the resulting data to the language context relation network.

7. The model generation system according to claim 1 , wherein
the processor section trains the language context relation network by acquiring the text line data for the training, converting the acquired text line data to a text line image through use of a predetermined font, inputting the resulting text line image to a predetermined visual feature extractor, and inputting the output of the predetermined visual feature extractor to the language context relation network.

8. The model generation system according to claim 1 , wherein
the existing labeled text line image is managed via the processor by a plurality of style-specific image groups formed by text line images of a same style, and
the processor section determines the variable of the visual feature extractor by training the text line recognition model through use of the labeled text line image in each of the style-specific image groups while the variable of the language context relation network is fixed at the determined variable.

9. A model generation method adopted by a model generation system for generating a text line recognition model that recognizes a text line included in a text line image,
the text line recognition model including
a visual feature extractor that, when executed by the model generation system, outputs image feature values from the text line image, and
a language context relation network that, when executed by the model generation system, inputs the feature values outputted from the visual feature extractor, and outputs the text line,

the model generation method comprising:
by the model generation system, the method including the following steps:
(1) determining a variable of the language context relation network by acquiring text data for training and thus training the language context relation network by using the acquired text data;
(2) determining a variable of the visual feature extractor by training the text line recognition model through use of an existing labeled text line image while the variable of the language context relation network is set according to step (1); and
(3) generating the text line recognition model while the variable of the language context relation network is set according to step (1) and the variable of the visual feature extractor is set according to step (2).

Assignees

• Hitachi Ltd

Inventors

• Kha Nguyen Cong
• Odate Ryosuke

Classifications

• G06V30/18 Primary

  Extraction of features or characteristics of the image · CPC title

• G06V30/262

  using context analysis, e.g. lexical, syntactic or semantic context · CPC title

• G06V30/19133

  Interactive pattern learning with a human teacher · CPC title

• G06V30/226

  of cursive writing · CPC title

• G06V10/82

  using neural networks · CPC title