KrutiExtract

A Universal Legacy Hindi PDF to Markdown Extractor

Extracting legacy Hindi fonts (like KrutiDev, Devlys, or Chanakya) from PDF documents has traditionally been a nightmare. Standard OCR and PDF text extractors read characters based on visual coordinates. Because Hindi matras (vowel marks like ि or े) overlap with consonants, this visual sorting scrambles the text (e.g., typing vkSj for 'और' gets extracted as vkjS, breaking the conversion into the nonsense word 'आरै').

KrutiExtract solves this problem by bypassing visual coordinate sorting and extracting the logical character stream directly from the PDF.

✨ Core Features

• Perfect Character Order: Preserves the original typist's keystrokes, completely avoiding spatial scrambling of matras.
• Smart Auto-Detect Engine: Automatically mathematically calculates text frequencies to detect whether the typist used KrutiDev / Devlys or Chanakya, instantly routing the text to the correct translation engine!
• English / Hinglish Bypass Mode: If the engine detects that the document contains modern Unicode Hindi, standard English, or Hinglish, it automatically bypasses legacy conversion entirely—perfectly preserving your modern texts!
• Dynamic English Preservation: Uses NLTK's English corpus to proactively detect and protect English words, numbers, acronyms, and HTML tags interspersed within mixed-font Hindi texts.
• Collision-Free Safe-Tokens: Markdown syntax elements (like # or *) clash with KrutiDev maps. KrutiExtract shields these structural elements using a unique $$$INDEX$$$ safe-token system during font conversion.

👁️ Advanced OCR & Image Cleaning

When the toolkit encounters text trapped inside a scanned image, it seamlessly hands it over to Tesseract OCR in the background.

• Native Hindi OCR: The toolkit hardcodes ocr_language="hin" into the internal pipeline. This permanently stops Tesseract from trying to hallucinate English letters onto Hindi image shapes, ensuring natively output Unicode Devanagari.
• OpenCV Image Cleaning (--enhance-ocr): Pass this flag to activate our built-in computer vision pipeline. The toolkit will automatically intercept blurry images in the PDF, upscale the resolution, apply unsharp masking, and run adaptive thresholding to perfectly reconstruct the text shapes before OCR scans it!

🚀 Installation

Ensure you have Python 3.8+ installed.

# Clone the repository
git clone https://github.com/epicbharat/krutiextract.git
cd krutiextract

# Create and activate a virtual environment (Recommended)
python -m venv venv
venv\Scripts\activate  # On Windows
# source venv/bin/activate  # On macOS/Linux

# Install dependencies
pip install -r requirements.txt

# Download the required NLTK English word corpus
python -c "import nltk; nltk.download('words')"

Note on Tesseract (Required for OCR)

If you are parsing scanned PDFs or PDFs with embedded images, you must have Tesseract OCR installed on your system.

• Windows: Download the installer and ensure you check the "Hindi" language pack during installation. Add Tesseract to your PATH.
• Linux: sudo apt-get install tesseract-ocr tesseract-ocr-hin

💻 CLI Usage Guide

KrutiExtract provides a powerful Command Line Interface (CLI).

1. Process a Single File

python -m src.cli --input "path/to/document.pdf" --output "path/to/output.md"

2. Batch Process a Directory

Recursively process an entire folder of PDFs:

python -m src.cli --dir "path/to/pdf/folder" --out-dir "path/to/markdown/folder"

3. Force a Specific Font Encoding

By default, the font is set to auto. You can force a specific bypass or mapping:

# Available options: krutidev, chanakya, auto, unicode, english
python -m src.cli --input "doc.pdf" --font chanakya

4. Enable OpenCV OCR Enhancement

Use this flag for old, blurry, or low-resolution scanned PDFs to artificially sharpen the text before extraction:

python -m src.cli --input "blurry_scan.pdf" --enhance-ocr

🤝 Contributing

Contributions are welcome! If you find a PDF that breaks the extraction logic or a rare character combination we missed, please open an issue and attach the sample PDF.

📄 License

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

Author: Bharat Choudhary (Lecturer, Department of Secondary Education, Govt. of Rajasthan)
Email: epicbharat@gmail.com
GitHub: https://github.com/epicbharat