Transcription extraction in Byzantine seals

This repository is related to the article

Théophile Rageau Laurence Likforman-Sulem, Attilio Fiandrotti, Victoria Eyharabide, Béatrice Caseau, Jean-Claude Cheynet, Character Recognition in Byzantine Seals with Deep Neural Networks, Digital Applications in Archaeology and Cultural Heritage, Volume 37, June 2025.

see also: https://github.com/LIK50/Byzantine_chars

Setup

repository

Clone the main repo:

git clone https://github.com/idsinge/ocrseals.git

Get the yolov5 submodule:

cd ocrseals
git submodule update --init

virtual environmnent

If you don't already have one, create a venv:

python3 -m venv venv
. ./venv/bin/activate

Then install pytorch with the appropriate method for your system by following the instructions at: https://pytorch.org/get-started/locally/ ; e.g.:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

Finally install requirements for yolov5 with:

pip install -r detection/yolov5/requirements.txt

Usage

The ocr_seals.py will apply detection, hough processing, classification, and transcript extraction for all images in data/input with a command like:

python3 ocr_seals.py --detection-model detection/detection_model.pt --classification-model classification/classification_model.pth --data data

The data directory will then contain:

• in data/detect the results of the detection process
• in data/houg the results of the hough process
• in data/trans the extracted transcriptions