Binarizing Severely Degraded Ancient Bamboo Slips: Dataset and Baseline


Chongsheng Zhang (Henan University, Ludwig-Maximilians-Universität München), Wanwan Fu (Henan University), Qilong Li (Henan Univeristy), SAMRA ZAFAR (Henan Univeristy), Zhanshuo Zhang (Henan Univeristy), Qiyan Li (Henan Univeristy), Gaojuan Fan (Henan Univeristy), Christian Heumann (Ludwig-Maximilians-Universität München)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Bamboo and wooden slips were the primary writing materials in China for more than 800 years, carrying valuable historical records. However, due to the lengthy corrosion and/or weathering effects over a period of two millennia, texts on the ancient bamboo (and wooden) slips typically suffer from severe degradation problems, such as ink deterioration and text blur, which render the binarization of severely degraded bamboo slip manuscripts a very challenging task. Due to the scarcity of benchmark dataset in this direction, in this work we aim to build HanBamboo, a specialized bamboo slips dataset with pixel-level annotations for research on binarization of severely degraded ancient manuscripts. HanBamboo comprises 1,000 infrared bamboo slip images displaying varying levels of ink degradation and text blur. Among them, bamboo slips exhibiting low-contrast ink traces characterized by significant fading and diminished visibility requires remarkably greater time in pixel-level annotation, indicating the inherent difficulty in binarizing these bamboo slips. As a minor contribution, we also propose a baseline approach MRA-Net, which is an Mamba-based Encoder-Decoder framework for degraded bamboo slips binarization that devises additional multi-scale discrete wavelet transform and adaptive reweighting attention mechanisms to capture and enhance the stroke details of the texts and suppress background noise during binarization. Comprehensive experiments on both HanBamboo and public document binarization benchmark datasets DIBCO 2017 and 2018 demonstrate the effectiveness of our baseline. Our dataset and code are available at: https://github.com/cszhangLMU/MRA-Net.

Citation

@inproceedings{Zhang_2025_BMVC,
author    = {Chongsheng Zhang and Wanwan Fu and Qilong Li and SAMRA ZAFAR and Zhanshuo Zhang and Qiyan Li and Gaojuan Fan and Christian Heumann},
title     = {Binarizing Severely Degraded Ancient Bamboo Slips: Dataset and Baseline},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_1177/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection