Year: 2026
Type: conference-paper
Abstract: This paper presents a complete workflow for the automatic transcription of Henri Pirenne's manuscripts, from corpus creation to HTR model evaluation.By combining pre-existing generic models with fine-tuning on Pirenne's writings, we achieve high recognition accuracy, reaching over 90% on the test set.Our transcription conventions ensure both fidelity to the manuscripts and reusability for other HTR projects.The resulting data and models provide a reproducible framework for future research, enabling large-scale analysis of Pirenne's work and contributing to the sustainable digital preservation of his work.
Institution: Université Libre de Bruxelles
Language: French
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7162977168