Robust Document OCR Preprocessing Pipeline

A robust preprocessing pipeline for document OCR that significantly improves Tesseract accuracy on mobile-captured ID documents.

🚀 Quick Start

Installation

# Install from PyPI
pip install RobustDocOCR

# Install with OCR support
pip install RobustDocOCR[ocr]

# Install with development dependencies
pip install RobustDocOCR[dev]

Basic Usage

from robustdococr import preprocess_document, load_image

# Load your document image
image = load_image("document.jpg")

# Apply preprocessing pipeline
results = preprocess_document(image, show_steps=True)

# Access preprocessed image
preprocessed_image = results['final']

Command Line Interface

# Process single image
robustdococr input.jpg --output output.jpg

# Process with intermediate steps display
robustdococr input.jpg --show-steps

📦 Features

4-Stage Preprocessing Pipeline

1. Deskewing: Straightens rotated documents using Hough transform
2. Binarization: Converts images to black & white using adaptive thresholding
3. Noise Removal: Cleans up artifacts using two-stage denoising
4. OCR Ready: Produces optimized images for Tesseract OCR

Key Technical Features

• Adaptive Thresholding: Handles varying lighting conditions (shadows, glare)
• Hough Transform Deskewing: Robust rotation correction (±45°)
• Two-Stage Denoising: Preserves text while removing artifacts
• 96% Text Retention: Minimal text loss during preprocessing
• Tesseract Optimized: Produces images ideal for OCR engines

🎯 Performance Metrics

Metric	Value
Text Retention Rate	96%
Character Improvement	+12%
Quality Distribution	85% Excellent, 12% Good, 3% Fair
Rotation Correction	Handles ±45° rotation effectively

📂 Project Structure

robustdococr/
 ├── preprocessing/          # Core preprocessing modules
 │   ├── deskewing.py        # Image straightening
 │   ├── binarization.py     # Adaptive thresholding
 │   ├── noise_removal.py    # Artifact cleaning
 │   └── pipeline.py         # Complete pipeline
 ├── utils/                  # Utility functions
 │   ├── image_utils.py      # Image utilities
 │   ├── ocr_utils.py        # OCR utilities
 │   └── visualization.py    # Visualization tools
 ├── cli.py                  # CLI entry point
 ├── main.py                 # Main module
 └── __init__.py             # Package initialization
tests/                      # Test suite
examples/                   # Example scripts
notebooks/                  # Jupyter notebooks
docs/                       # Documentation

🔧 Configuration

Requirements

• Python 3.8+
• OpenCV
• NumPy
• Pillow
• Matplotlib (for visualization)
• Tesseract OCR (optional, for OCR features)

Installation Options

# Basic installation
pip install RobustDocOCR

# Development installation (includes test and dev dependencies)
pip install RobustDocOCR[dev]

# Installation with OCR support
pip install RobustDocOCR[ocr]

# Installation with all extras
pip install RobustDocOCR[all]

📊 Technical Specifications

Deskewing Algorithm

• Edge Detection: Canny edge detector with thresholds (50, 150)
• Line Detection: Hough Line Transform with threshold 200
• Angle Calculation: Median angle from detected lines for robustness
• Rotation: Affine transformation with cubic interpolation

Binarization Algorithm

• CLAHE Enhancement: Contrast Limited Adaptive Histogram Equalization
  • clipLimit: 2.0
  • tileGridSize: (8, 8)
• Adaptive Thresholding: Gaussian-weighted local thresholding
  • blockSize: 25
  • C: 10
  • Method: ADAPTIVE_THRESH_GAUSSIAN_C

Noise Removal Algorithm

• Stage 1: Non-Local Means Denoising (h=10) applied before binarization
• Stage 2: Morphological operations (2×2 kernel, 1 iteration) applied after binarization

🧪 Testing

Run the test suite:

pytest

Run tests with coverage:

pytest --cov=robustdococr --cov-report=html

📚 Documentation

• Architecture Documentation
• Usage Guide
• Decision Log
• API Reference
• Kaggle Notebook - Complete preprocessing pipeline demonstration

🤝 Contributing

We welcome contributions! Please see our:

• Contributing Guidelines
• Code of Conduct
• Issue Templates

📄 License

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

🎓 Citation

If you use this pipeline in your research, please cite:

@misc{robust-doc-ocr-preprocessing,
  author = {3BSALAM},
  title = {Robust Document OCR Preprocessing Pipeline},
  year = {2026},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/3bsalam-1/RobustDocOCR}}
}

🔗 Related Projects

• Tesseract OCR
• OpenCV
• MIDV-500 Dataset

📦 PyPI

This package is available on PyPI: https://pypi.org/project/RobustDocOCR/

---

© 2026 Robust Document OCR Preprocessing Pipeline