Scholarlink

<p>Ancient Manchu word recognition is a low-resource heritage OCR problem shaped by degraded document images, connected vertical script, scarce annotations, and limited expert knowledge. This Review synthesizes methodological progress from statistical and rule-based approaches to feature-engineered machine learning, deep learning, transfer learning, and vision-language recognition. It highlights persistent benchmark, robustness, and cross-collection generalization bottlenecks and outlines an expert-guided roadmap for reliable heritage OCR.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A review of ancient Manchu word recognition for low-resource heritage OCR

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月27日

查看全文
  • Haipeng Sun,
  • Xiaojun Bi,
  • Haoran Li,
  • Weizheng Qiao,
  • Zheng Chen

摘要

Ancient Manchu word recognition is a low-resource heritage OCR problem shaped by degraded document images, connected vertical script, scarce annotations, and limited expert knowledge. This Review synthesizes methodological progress from statistical and rule-based approaches to feature-engineered machine learning, deep learning, transfer learning, and vision-language recognition. It highlights persistent benchmark, robustness, and cross-collection generalization bottlenecks and outlines an expert-guided roadmap for reliable heritage OCR.