Historical Text Recognition (RenAIssance Project)

This project implements an OCR (Optical Character Recognition) system specifically designed for historical text recognition, focusing on early modern printed sources. The system uses a combination of deep learning architectures to accurately transcribe historical documents while handling various challenges like layout variations and text embellishments.

Project Structure

.
├── data/               # Dataset storage
│   ├── raw/           # Original PDF scans
│   ├── processed/     # Processed images and annotations
│   └── transcriptions/# Reference transcriptions
├── notebooks/         # Jupyter notebooks for analysis
├── src/              # Source code
│   ├── data/         # Data processing utilities
│   ├── models/       # Model architectures
│   ├── training/     # Training scripts
│   └── utils/        # Utility functions
├── outputs/          # Model outputs and results
└── requirements.txt  # Project dependencies

Setup

1. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

Model Architecture

The OCR system uses a hybrid architecture combining:

• A CNN backbone for feature extraction
• A Transformer encoder for sequence modeling
• A CTC decoder for text recognition

Key features:

• Layout-aware text detection
• Robust handling of historical fonts and styles
• Support for early modern English text variations

Evaluation Metrics

The model is evaluated using:

• Character Error Rate (CER)
• Word Error Rate (WER)
• Layout detection accuracy
• Processing speed

Usage

1. Data Preparation:

python src/data/prepare_data.py

2. Training:

python src/training/train.py

3. Inference:

python src/inference.py

Results

Model performance metrics and visualizations are stored in the outputs/ directory.

License

This project is licensed under the MIT License - see the LICENSE file for details.