Image processing apparatus, control method thereof, and storage medium

Patent metadata

Field	Value
Publication number	US-12586400-B2
Application number	US-202217837564-A
Country	US
Kind code	B2
Filing date	Jun 10, 2022
Priority date	Jun 16, 2021
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

A training image that simulates a character block entered by part of a character in another row is generated efficiently. A noise is added in the vicinity of an end portion of a character image so that a noise that seems to be caused by entering of part of a character in another row is reproduced for the character image representing a handwritten character.

First claim

What is claimed is:

1 . An image processing apparatus comprising:
a memory that stores a program; and a processor that executes the program to perform:
superimposing a noise cut out from another character image in a vicinity of an end portion within a character image representing a handwritten character, wherein the noise is not superimposed in other areas within the character image than the vicinity of the end portion within the character image, wherein
in the superimposing, in a case where a noise addition position is a top, the noise is cut out from a bottom end of the character image, and, in a case where the noise addition position is a bottom, the noise is cut out from a top end of the character image; and
generating training data by associating a character image in which the noise is superimposed and a correct answer class with each other, wherein
the generated training data is used for training of a model of a neural network to ignore the noise and reduce erroneous recognition of the handwritten character.

2 . The image processing apparatus according to claim 1 , wherein
in the superimposing:
processing to extend a frame of the character image is performed; and
the noise is superimposed in a vicinity of an end portion within the frame-extended character image, wherein the noise is not superimposed in other areas within the frame-extended character image than the vicinity of the end portion within the frame-extended character image.

3 . The image processing apparatus according to claim 2 , wherein
in the superimposing, a frame is extended in one of upward, downward, leftward, and rightward directions and the noise is superimposed in a vicinity of an end portion in the direction in which the frame is extended, wherein the noise is not superimposed in other areas within the frame-extended character image than the vicinity of the end portion in the direction in which the frame is extended.

4 . The image processing apparatus according to claim 1 , wherein
in the superimposing, in a case where the noise is superimposed in the vicinity of the end portion within the character image, the noise is superimposed so that the noise comes into contact with the end portion.

5 . The image processing apparatus according to claim 1 , wherein
in the superimposing, the noise is superimposed so that a character pixel corresponding to a handwritten character within the character image and a noise pixel corresponding to the noise come into contact with each other.

6 . The image processing apparatus according to claim 1 , wherein
the processor executes the program to perform:
obtaining a rectangular area corresponding to a character described in a document from a scanned image obtained by scanning the document;
separating, in a case where the rectangular area corresponds to a plurality of character rows described in the document, the rectangular area into a rectangular area for each character row; and
outputting character recognition results for the rectangular area by estimation using the model which is trained by using the generated training data, wherein
in the estimating:
for the rectangular area separated by the separating, the estimation is performed by the model trained by using the generated training data; and
for the rectangular area not separated by the separating, the estimation is performed by a trained model which is trained using training data in which the character image to which no noise is superimposed in the superimposing and a correct answer class are associated with each other.

7 . The image processing apparatus according to claim 6 , wherein
the processor executes the program to perform training the model of the neural network by using the generated training data.

8 . A control method of an image processing apparatus, the control method comprising:
superimposing a noise cut out from another character image in a vicinity of an end portion within a character image representing a handwritten character, wherein the noise is not superimposed in other areas within the character image than the vicinity of the end portion within the character image, wherein in the superimposing, in a case where a noise addition position is a top, the noise is cut out from a bottom end of the character image, and, in a case where the noise addition position is a bottom, the noise is cut out from a top end of the character image; and generating training data by associating a character image in which the noise is superimposed and a correct answer class with each other, wherein the generated training data is used for training of a model of a neural network to ignore the noise and reduce erroneous recognition of the handwritten character.

9 . A non-transitory computer readable storage medium storing a program for causing a computer to perform a control method of an image processing apparatus, the control method comprising:
superimposing a noise cut out from another character image in a vicinity of an end portion within a character image representing a handwritten character, wherein the noise is not superimposed in other areas within the character image than the vicinity of the end portion within the character image, wherein in the superimposing, in a case where a noise addition position is a top, the noise is cut out from a bottom end of the character image, and, in a case where the noise addition position is a bottom, the noise is cut out from a top end of the character image; and generating training data by associating a character image in which the noise is superimposed and a correct answer class with each other, wherein the generated training data is used for training of a model of a neural network to ignore the noise and reduce erroneous recognition of the handwritten character.

10 . The image processing apparatus according to claim 1 , wherein
in the superimposing, in a case where a noise addition position is a left, the noise is cut out from a right end of the character image, and, in a case where the noise addition position is a right, the noise is cut out from a left end of the character image.

Assignees

• Canon Kk

Inventors

• Seto Hidekazu
• Okuma Keisui
• Maeda Yusuke

Classifications

• G06V30/22

  characterised by the type of writing · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V30/19147 Primary

  Obtaining sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title