Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
SYSTEM FOR TRAINING MACHINE LEARNING MODEL RECOGNIZING CHARACTER OF TEXT IMAGE
Document Type and Number:
Japanese Patent JP2022161564
Kind Code:
A
Abstract:
To improve recognition precision of a text image recognition model with less training data.SOLUTION: A system stores a machine learning model which recognizes characters in a text image. The machine learning model which recognizes characters in a text image includes: a character segmentation network which extracts a visual feature quantity from the text image and generates a character border box from the text image; a domain adaptation network which classifies a domain of the text image based upon the visual feature quantity; and a text recognition network which recognizes the characters in the text image based upon the character border box and visual feature quantity. The system inverts a gradient of training of the domain adaptation network to a negative gradient and propagates the same reversely in the character segmentation network. The system propagates the gradient of the training of the text recognition network reversely in the character segmentation network.SELECTED DRAWING: Figure 1

Inventors:
NGUYEN CONGKHA
ODATE RYOSUKE
Application Number:
JP2021066477A
Publication Date:
October 21, 2022
Filing Date:
April 09, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
HITACHI LTD
International Classes:
G06V30/194; G06N3/08; G06T7/00
Attorney, Agent or Firm:
Fujio Patent Attorneys Corporation