A multi-modal adaptive recognition, image restoration and domain knowledge verification method and system for archives of the Republic of China

The multimodal adaptive recognition system has solved the problem of identifying and restoring Republican-era archives, achieved efficient automated processing, improved the full-text retrieval capability of Republican-era archives, and reduced manual processing time and health losses.

CN122290147APending Publication Date: 2026-06-26胡勇
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
胡勇
Filing Date
2026-04-13
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing technologies struggle to identify handwritten cursive script from the Republican era, restore noise in mimeographed documents, and understand the context of Republican-era language. This makes it impossible to conduct full-text searches of Republican-era archives, requiring researchers to manually transcribe them, which is inefficient and harmful to their health.

Method used

A multimodal adaptive recognition system is adopted, combined with conditional generative adversarial networks for image restoration, and a handwritten font library and knowledge graph of famous figures from the Republic of China are constructed to realize handwriting style detection and semantic verification, and support human-computer collaborative proofreading and feedback learning.

Benefits of technology

It has achieved high-precision automatic identification and structured conversion of Republican-era archives, improving processing efficiency. The processing time for a single page of archives has been reduced from 20 minutes to 3 minutes, and the system continues to optimize itself during use.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122290147A_ABST
    Figure CN122290147A_ABST
Patent Text Reader

Abstract

This invention provides a multimodal adaptive recognition, image restoration, and domain knowledge verification method and system for Republican-era archives, relating to the field of text and image optical processing technology. The system includes an acquisition unit, a preprocessing unit, a recognition unit, a knowledge graph storage unit, a verification unit, an interaction unit, a feedback learning unit, a central processing unit, and a storage unit. The acquisition unit is used to acquire original images of Republican-era archives and supports integration with scanners, document scanners, and import of local image files. This invention solves the technical problems of existing technologies, such as difficulty in recognizing Republican-era cursive handwriting, difficulty in restoring noise from mimeographed documents, and difficulty in understanding the context of Republican-era language. It achieves intelligent and high-precision conversion from archival images to searchable structured text, transforming the work mode of archivists from manual copying to intelligent verification, thereby significantly improving work efficiency.
Need to check novelty before this filing date? Find Prior Art