CTC Transcription Alignment of the Bullinger Letters: Automatic Improvement of Annotation Quality
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918122101932032 |
|---|---|
| author | Peer, Marco Scius-Bertrand, Anna Fischer, Andreas |
| author_facet | Peer, Marco Scius-Bertrand, Anna Fischer, Andreas |
| contents | Handwritten text recognition for historical documents remains challenging due to handwriting variability, degraded sources, and limited layout-aware annotations. In this work, we address annotation errors - particularly hyphenation issues - in the Bullinger correspondence, a large 16th-century letter collection. We introduce a self-training method based on a CTC alignment algorithm that matches full transcriptions to text line images using dynamic programming and model output probabilities trained with the CTC loss. Our approach improves performance (e.g., by 1.1 percentage points CER with PyLaia) and increases alignment accuracy. Interestingly, we find that weaker models yield more accurate alignments, enabling an iterative training strategy. We release a new manually corrected subset of 100 pages from the Bullinger dataset, along with our code and benchmarks. Our approach can be applied iteratively to further improve the CER as well as the alignment quality for text recognition pipelines. Code and data are available via https://github.com/andreas-fischer-unifr/nntp. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_07904 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | CTC Transcription Alignment of the Bullinger Letters: Automatic Improvement of Annotation Quality Peer, Marco Scius-Bertrand, Anna Fischer, Andreas Computer Vision and Pattern Recognition Handwritten text recognition for historical documents remains challenging due to handwriting variability, degraded sources, and limited layout-aware annotations. In this work, we address annotation errors - particularly hyphenation issues - in the Bullinger correspondence, a large 16th-century letter collection. We introduce a self-training method based on a CTC alignment algorithm that matches full transcriptions to text line images using dynamic programming and model output probabilities trained with the CTC loss. Our approach improves performance (e.g., by 1.1 percentage points CER with PyLaia) and increases alignment accuracy. Interestingly, we find that weaker models yield more accurate alignments, enabling an iterative training strategy. We release a new manually corrected subset of 100 pages from the Bullinger dataset, along with our code and benchmarks. Our approach can be applied iteratively to further improve the CER as well as the alignment quality for text recognition pipelines. Code and data are available via https://github.com/andreas-fischer-unifr/nntp. |
| title | CTC Transcription Alignment of the Bullinger Letters: Automatic Improvement of Annotation Quality |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2508.07904 |