Character Recognition 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.

Classification

This repository demonstrates the inference part of the classification step described in the article.

The images directory contains 2224 cropped characters (28 classes), including the background classe bg.

The classes are:

• C moon-shaped sigma
• CT ligature
• Croisette
• R = Beta
• S = καί
• V = Y
• bg
• ligature OU
• sigma
• Α
• Β
• Γ
• Δ
• Ε
• Ζ
• Η
• Θ
• Ι
• Κ
• Λ
• Μ
• Ν
• Ξ
• Ο
• Π
• Ρ = rho
• Τ
• Φ
• Χ
• ω
• И = N

Usage

You can see a demonstration of the classification by using the read_seals.py:

# install requirements
pip install -r requirements.txt

# run the classifier
python3 read_seals.py

You can change on which image the classifier is run by editing the read_seals.py script.

Full inference pipeline

To demonstrate the whole pipeline of inference consisting of the 3 steps (detection, classification, transcription) that allows going from an image of a seal to the extracted character sequence, another repository has been set-up here:

https://github.com/idsinge/ocrseals