Title:
LEARNING MODEL, LEARNING MODEL GENERATION METHOD, ROTATION CHARACTER IDENTIFICATION DEVICE, ROTATION CHARACTER IDENTIFICATION METHOD, AND COMPUTER PROGRAM
Document Type and Number:
Japanese Patent JP2023085076
Kind Code:
A
Abstract:
To provide a learning model capable of identifying a rotation character with high accuracy.SOLUTION: A learning model includes: an input layer that accepts input of character information including at least a character, a rotation character image, which is an image representing a rotation character, which is a rotation character obtained by rotating the character at a predetermined angle about a symmetrical point, and rotation angle information with a predetermined angle; an output layer that identifies, when any rotation character image is input, the character included in the rotation character image, calculates a character identification rate, and outputs the highest identification rate value, character information corresponding to the value, and rotation angle information; and an intermediate layer where a parameter is learned on the basis of the character information, the rotation characteristic image, and the rotation information. The character is obtained by excluding a plurality of characters that has identical or similar rotation symmetry to each other, and is operated, when an arbitrary rotation character image is input to the input layer, by the intermediate layer, and causes a computer to output the identification rate, the character information, and the rotation angle information from the output layer.SELECTED DRAWING: Figure 4
Inventors:
MITA ATSUSHI
KAI SUSUMU
KAI SUSUMU
Application Number:
JP2021199558A
Publication Date:
June 20, 2023
Filing Date:
December 08, 2021
Export Citation:
Assignee:
UNITEX KK
International Classes:
G06V30/146; G06T7/00
Attorney, Agent or Firm:
Masaru Takenouchi
Yoshinori Yokobori
Yoshinori Yokobori