text-detector-bot

A Telegram bot and Flask API for detecting and recognizing text in images using PaddleOCR and TrOCR.

Features

• Detects text regions in images using PaddleOCR.
• Recognizes text in detected regions using TrOCR (transformer-based OCR).
• REST API for programmatic access.
• Telegram bot for easy image-to-text interaction.

---

Requirements

• Python 3.7+
• pip

Python Dependencies

Install all required packages:

pip install paddleocr opencv-python-headless numpy pillow matplotlib flask python-telegram-bot transformers torch

Note:

• paddleocr may require additional system dependencies (see PaddleOCR installation guide).
• torch version should match your CUDA version if using GPU.

---

Setup

1. Clone the repository

git clone <repo-url>
cd text-detector-bot

2. Configure Telegram Bot

Create a file named private_data.json in the project root with the following content:

{
  "telegram_token": "YOUR_TELEGRAM_BOT_TOKEN"
}

Get your bot token from BotFather on Telegram.

---

Usage

1. Start the Flask API Server

python flask_server.py

• The server will run on http://127.0.0.1:5000.
• Main endpoint: POST /api/process_image/
  • Request JSON: { "image": "<base64-encoded-image>" }
  • Response JSON: { "bboxes": [...], "binary_image": "<base64>", "text": ["..."] }

2. Start the Telegram Bot

python telegram_bot.py

• The bot will listen for image messages.
• When you send an image, it will reply with:
  • The image with detected text regions highlighted.
  • The binarized (preprocessed) image.
  • The recognized text.

---

How it Works

• Detection: PaddleOCR detects text regions in the image.
• Recognition: Each detected region is cropped and passed to TrOCR for text recognition.
• API: The Flask server exposes an endpoint for image processing.
• Telegram Bot: Forwards images to the API and returns results to the user.

---

File Structure

• text_detector.py — PaddleOCR-based text detection.
• trocr.py — TrOCR-based text recognition.
• flask_server.py — REST API server.
• telegram_bot.py — Telegram bot logic.
• utils.py — Image conversion utilities.

---

Notes

• The Telegram bot expects the Flask server to be running locally at http://127.0.0.1:5000.
• For production, consider deploying the Flask server and bot separately, and update the API URL in telegram_bot.py if needed.
• For GPU acceleration, ensure CUDA is installed and torch is installed with CUDA support.

---

Acknowledgements

• PaddleOCR
• TrOCR (Microsoft)
• python-telegram-bot