Granted Patent B2

US 12,406,515 · App. 17/884,825 · Granted Sep 2, 2025

Inventors: Go Oishi (Kanagawa, JP); Yusuke Muramatsu (Tokyo, JP)

Assignee: CANON KABUSHIKI KAISHA

G06V30/333G06T5/50G06V30/164G06V30/19147

Abstract

Claims (33)

1. An information processing apparatus comprising:

at least one memory that stores a program; and

at least one processor that executes the program to:

acquire a handwriting image;

generate a correct label image by binarizing the acquired handwriting image, the correct label image being indicative of handwriting pixels in the acquired handwriting image;

obtain a first scan image generated by reading a blank sheet;

acquire a noise image by cutting a part of the obtained first scan image out, a position of the part being randomly determined from the obtained first scan image;

synthesize the acquired handwriting image and the acquired noise image to generate a synthesized image; and

use the synthesized image and the correct label image as learning data for generating a learning model that is usable for estimating handwriting pixels from an image.

2. The information processing apparatus according to claim 1 , wherein the handwriting image is obtained by performing at least one of a rotation processing, a cutting out processing, a magnification varying processing, and a brightness changing processing for a second scan image generated by reading a sheet including handwriting.

3. The information processing apparatus according to claim 1 , wherein a pixel value of the synthesized image is determined by comparing pixel values of corresponding pixels in the acquired handwriting image and the acquired noise image.

4. The information processing apparatus according to claim 1 , wherein the synthesized image is generated by synthesizing the acquired handwriting image, the acquired noise image, and an image including information associated with a background.

5. The information processing apparatus according to claim 1 , wherein the at least one processor executes the program to further:

generate the learning model for estimating handwriting pixels from an image by using the learning data.

6. The information processing apparatus according to claim 5 , wherein the at least one processor executes the program to further:

estimate handwriting pixels from a process target image by using the generated learning model.

7. The information processing apparatus according to claim 6 , wherein the at least one processor executes the program to further:

generate, based on a result of the estimating, an extracted image corresponding to handwriting and an extracted image corresponding to a background from the process target image.

8. The information processing apparatus according to claim 5 , wherein the handwriting pixels are estimated from a process target image by an extraction processing apparatus that is different from the information processing apparatus and uses the generated learning model for the estimation.

9. An information processing method comprising:

acquiring a handwriting image;

generating a correct label image by binarizing the acquired handwriting image, the correct label image being indicative of handwriting pixels in the acquired handwriting image;

obtaining a first scan image generated by reading a blank sheet;

acquiring a noise image by cutting a part of the obtained first scan image out, a position of the part being randomly determined from the obtained first scan image;

synthesizing the acquired handwriting image and the acquired noise image to generate a synthesized image; and

using the synthesized image and the correct label image as learning data for generating a learning model that is usable for estimating handwriting pixels from an image.

10. A non-transitory computer readable storage medium storing instructions that causes a computer to:

acquire a handwriting image;

generate a correct label image by binarizing the acquired handwriting image, the correct label image being indicative of handwriting pixels in the acquired handwriting image;

obtain a first scan image generated by reading a blank sheet;

acquire a noise image by cutting a part of the obtained first scan image out, a position of the part being randomly determined from the obtained first scan image;

synthesize the acquired handwriting image and the acquired noise image to generate a synthesized image; and

use the synthesized image and the correct label image as learning data for generating a learning model that is usable for estimating handwriting pixels from an image.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (19)