HTR ground-truth of the CHI-KNOW-PO project.
The CHI-KNOW-PO project aims to digitize a corpus of poetic anthologies, commentaries, dictionaries and encyclopedias from the Chinese medieval period (ca. 200-1000) and process them using HTR.
Official page of the htr project
Documentation of the research project
To date, dataset contains 327 images, for a total of:
- 1.248 TextRegions
- 13.759 TextLines
- 104.536 Glyphs
The dataset (images + XML) is available on Zenodo.
TBD
We provide for each image a pageXML file containing three level of information:
- TextRegion localisation, with a semantic tag (e.g.
MainText), following the SegmOnto ontology; - Baseline localisation and surrounding polygon of the line. Each baseline contains a semantic tag;
- Text.
<TextRegion id="79718" custom="structure {type:MainText;}">
<Coords points="2575,4313 2575,2861 2563,1405 219,1413 216,4307 2575,4313"/>
<TextLine id="870481" custom="structure {type:Text;}">
<Coords points="2491,1414 2414,1414 2414,1515 2392,1534 2411,1584 2392,1627 2411,1679 2397,1732 2414,1751 2397,2082 2417,2102 2400,2178 2419,2206 2400,2258 2425,2338 2571,2352 2559,1411 2491,1414"/>
<Baseline points="2492,1415 2504,2327"/>
<TextEquiv>
<Unicode>見惡焉其終也已</Unicode>
</TextEquiv>
</TextLine>
<TextLine id="870482" custom="structure {type:Commentary;}">
<Coords points="2535,2356 2488,2356 2477,2395 2493,2507 2474,2545 2493,2595 2474,2732 2491,2822 2477,2852 2493,2998 2480,3025 2493,3132 2482,3250 2502,3266 2641,3258 2630,2354 2535,2356"/>
<Baseline points="2536,2359 2552,3267"/>
<TextEquiv>
<Unicode>○今案見論語陽</Unicode>
</TextEquiv>
</TextLine>Annotations have been made on the Calfa Vision platform, a free web-based annotation tool for documents and images designed for Oriental scripts.
For HTR, we have first train a generic model with all the data, using Calfa Vision platform, then (re)fine-tuned this generic model with the targeted manuscript. On a new in-domain test set, we get the following results:
| Manuscript | N° | Accuracy |
|---|---|---|
| Li Wenxuan | A-1 | 99.38 (± 1.2) |
| Liuchen Wenxuan | A-2 | 98.84 (± 1.8) |
| Yutai | A-3 | 98.52 (± 1.2) |
| Tangshi | A-4 | 99.25 (± 1.8) |
| Beitang | S-1 | 98.76 (± 1.8) |
| Bowu zhi | S-2 | 99.18 (± 1.8) |
| Chuxue | S-3 | 97.57 (± 1.7) |
| Erya | S-4 | 96.57 (± 0.4) |
| Maoshi shu | S-5 | 98.42 (± 1.8) |
| Yiwen | S-6 | 98.72 (± 1.7) |
| Zhibuzu | S7 | 98.70 (± 1.8) |
| Shiwen leiju | T-1 | 97.47 (± 4.5) |
| Qimin yaoshu | T-2 | 99.35 (± 2.8) |
| Xinzhai | T-3 | 97.61 (± 3.2) |
For the reading order, we have defined a three step approach, combining a local sorting and a global sorting, achieving a 97.81% of accuracy.
@InProceedings{10.1007/978-3-031-70642-4_3,
author="Bizais-Lillig, Marie
and Vidal-Gor{\`e}ne, Chahan
and Dupin, Boris",
editor="Mouch{\`e}re, Harold
and Zhu, Anna",
title="Optimizing HTR and Reading Order Strategies for Chinese Imperial Editions with Few-Shot Learning",
booktitle="Document Analysis and Recognition -- ICDAR 2024 Workshops",
year="2024",
publisher="Springer Nature Switzerland",
address="Cham",
pages="37--56"
}