Reproduce “Joint Layout Analysis, Character Detection and Recognition for Historical Document Digitization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weihong Ma, Hesuo Zhang, Lianwen Jin, Sihang Wu, Jiapeng Wang, Yongpan Wang
Researchers from South China University of Technology and Alibaba Group developed an end-to-end framework that simultaneously performs layout analysis, character detection, and recognition for historical documents. The system achieved a 98.60% F-score for layout analysis, a 96.03% H-mean for text line detection (IoU=0.7), and a 96.07% Correct Rate in overall document restoration, further providing the MTHv2 dataset.
Key insights
- Integrating layout analysis, character detection, and recognition into a single end-to-end framework enables synergistic learning and reduces error propagation.
- A character-based approach for text line grouping, guided by layout information, offers superior robustness in handling complex and varied document column structures.
- Utilizing an auxiliary text line recognition model as a re-score mechanism effectively improves recognition accuracy by injecting contextual language information for ambiguous characters.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.