📝 Write2Text

Handwritten Text to Printed Text Conversion using Machine Learning

---

📋 Description

Write2Text converts images containing handwritten text into printed text. The system segments text into individual words and recognizes them using the TrOCR model.

🔄 How It Works

1. Image Upload → User uploads an image with handwritten text
2. Segmentation → System divides text into separate words and rows
3. Recognition → Each word is processed by the OCR model (TrOCR)
4. Result → Recognized text is returned in printed format

---

✨ Features

• 🖼️ Image processing with various formats support
• ✂️ Intelligent text segmentation into words
• 🤖 ML-based recognition using TrOCR
• 🌐 Web interface for easy usage
• 🚀 REST API for integration
• 💾 Local model storage for offline work

---

🚀 Quick Start

Installation

# Clone repository
git clone <repository-url>
cd write2text

# Create virtual environment
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # Linux/Mac

# Install dependencies
pip install -r requirements.txt

# Download OCR model
python download_model.py

Usage

# Start server
python main.py

Open frontend/index.html or navigate to http://localhost:8000

---

📁 Project Structure

write2text/
├── backend/          # FastAPI application
├── frontend/         # Web interface
├── ml/               # ML components
│   ├── preprocessing/  # Text segmentation
│   ├── ocr/          # OCR model
│   └── notebooks/    # Experiments
├── models/           # Saved ML models
├── main.py           # Entry point
└── download_model.py # Model download script

---

🛠️ Technologies

• FastAPI — Web framework
• PyTorch — Deep learning
• Transformers — Pre-trained models
• TrOCR — Text recognition
• OpenCV — Image processing

---

📡 API

POST /segment

Upload image for processing.

curl -X POST "http://localhost:8000/segment" \
     -F "file=@image.png"

Response:

{
  "status": "success",
  "word_count": 42,
  "recognized_text": "Recognized text...",
  "word_frames": [...]
}

---

🔧 Configuration

Change OCR model in ml/ocr/model.py:

ocr_model = OCRModel(model_name)

Device is auto-detected: CUDA → MPS → CPU

---

🙏 Acknowledgments

• TrOCR by Microsoft
• raxtemur/trocr-base-ru for Russian language

---