FP-THD: Full page transcription of historical documents

TL;DR

The paper presents FP-THD, a full-page transcription pipeline for medieval Latin documents that preserves original characters, abbreviations, and diacritics by combining a layout-analysis stage (ParseNet) with an extended masked autoencoder OCR (MAE-ViT). The system outputs PAGE XML layout plus transcriptions in Markdown and TXT formats, enabling faithful digitization and easier scholarly revision. Evaluations on handwritten (Rodrigo, Bentham) and printed (Molino) datasets show competitive CER and WER without post-processing, highlighting strong cross-domain generalization and preservation of historical glyphs. This work advances digital humanities workflows by delivering an end-to-end, open-tool pipeline for precise historical Latin transcription and lays groundwork for future Latin abbreviation expansion and language translation tasks using large language models.

Abstract

The transcription of historical documents written in Latin in XV and XVI centuries has special challenges as it must maintain the characters and special symbols that have distinct meanings to ensure that historical texts retain their original style and significance. This work proposes a pipeline for the transcription of historical documents preserving these special features. We propose to extend an existing text line recognition method with a layout analysis model. We analyze historical text images using a layout analysis model to extract text lines, which are then processed by an OCR model to generate a fully digitized page. We showed that our pipeline facilitates the processing of the page and produces an efficient result. We evaluated our approach on multiple datasets and demonstrate that the masked autoencoder effectively processes different types of text, including handwritten, printed and multi-language.

Figures (2)

• Figure 1: FP-THD architecture Overview: Layout Analysis and Masked Auto-encoder with Vision Transformer
• Figure 2: Example text lines by datasets.