OCR Service

This is a stand alone character recognition microservice that can be accessed via a REST API.

The service supports two OCR engines:

• Tesseract: Fast and efficient for standard document OCR
• PaddleOCR: Better for multi-directional and rotated text (default)

Documentation

https://gunthercox.com/ocr-service/

Docker Hub Image

https://hub.docker.com/r/gunthercox/ocr-service

Available Image Variants

The service provides three Docker image variants to suit different deployment needs:

1. Combined (default) - Both engines available

   docker pull gunthercox/ocr-service:latest
   docker pull gunthercox/ocr-service:1.0.0

   • Python 3.12
   • Includes both Tesseract and PaddleOCR
   • Largest image size (includes 140+ Tesseract language packs)
   • Use when you need flexibility to choose engines per request

2. Tesseract only - Smaller image for document OCR

   docker pull gunthercox/ocr-service:latest-tesseract
   docker pull gunthercox/ocr-service:1.0.0-tesseract

   • Python 3.12
   • Includes only Tesseract with all language packs
   • Use for standard document OCR workloads

3. PaddleOCR only - Smallest image, newest Python

   docker pull gunthercox/ocr-service:latest-paddleocr
   docker pull gunthercox/ocr-service:1.0.0-paddleocr

   • Python 3.13
   • Includes only PaddleOCR (no Tesseract language packs)
   • Smallest image size
   • Use for rotated/multi-directional text or when image size matters

Note: Engine-specific images will return an error if you request an unavailable engine. For example, the Tesseract-only image will reject requests when engine=paddleocr is specified in the request body.

API Usage

The service expects a multipart/form-data request with the following fields:

• image (required): The image file to process
• engine (optional): 'tesseract' or 'paddleocr' (default: 'paddleocr')
• lang (optional): Language code (format depends on engine)

Example Request

# Using default engine (PaddleOCR)
curl -X POST -F "image=@example.png" http://localhost:5000/

# Using Tesseract engine
curl -X POST -F "image=@example.png" -F "engine=tesseract" -F "lang=eng" http://localhost:5000/

# Using PaddleOCR with Chinese
curl -X POST -F "image=@example.png" -F "engine=paddleocr" -F "lang=ch" http://localhost:5000/

Response Format

Data will be returned in the following format:

{
    "text": "Extracted text from the image",
    "regions": [
        {
            "bbox": [[10, 20], [100, 20], [100, 40], [10, 40]],
            "text": "Detected text",
            "confidence": 0.95
        }
    ]
}