Mistral OCR Processing Tool

A versatile command-line tool for processing PDFs and images with Mistral's OCR API. This script extracts text content as markdown and preserves any embedded images.

Features

• Multi-format Support: Process both PDF documents and image files (JPG, PNG, GIF, BMP, TIFF, WEBP)
• Text Extraction: Convert documents and images to clean markdown
• Image Preservation: Extract embedded images from documents if they are available in array
• Batch Processing: Process all supported files in a directory automatically
• Structured Output: Creates markdown files with the same name as the input files

Requirements

• Python 3.6+
• mistralai Python package
• Mistral API key

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/mistralocr.git
   cd mistralocr

2. Install the required dependencies:

   pip install mistralai

3. Set up your Mistral API key:

   • Option 1: Edit the script and replace "your_api_key_here" with your actual API key
   • Option 2: Set an environment variable:

     export MISTRAL_API_KEY="your_api_key_here"

Usage

1. Place the script in a directory containing PDF and/or image files you want to process
2. Run the script:

   python mistral_ocr.py
3. The script will:
   • Find all supported files in the current directory
   • Process each file with the Mistral OCR API
   • Save extracted text as markdown files (.md)
   • Extract embedded images to {filename}_images/ folders
   • Save full JSON responses as {filename}_full.json files

Output Files

For each input file example.pdf or example.jpg, the script creates:

• example.md: The extracted text content in markdown format
• example_full.json: The complete JSON response from the Mistral OCR API
• example_images/ (if images are found): A directory containing any extracted images

How It Works

1. File Detection: The script scans the current directory for PDF and image files
2. Upload: Files are uploaded to Mistral's API
3. OCR Processing: The API extracts text and identifies any embedded images
4. Content Extraction: The script parses the API response to get markdown content
5. Image Extraction: Any embedded images are saved to a separate directory
6. Output Generation: Markdown files are created with references to the extracted images

API Response Structure

The Mistral OCR API returns a JSON structure with the following key elements:

{
  "pages": [
    {
      "index": 0,
      "markdown": "# Document Title\n\nSample text content with ![img-0.jpeg](img-0.jpeg) image reference.",
      "images": [
        {
          "id": "img-0.jpeg",
          "top_left_x": 253,
          "top_left_y": 473,
          "bottom_right_x": 1630,
          "bottom_right_y": 1792,
          "image_base64": "[BASE64_DATA_REMOVED]",
          "format": "jpeg"
        }
      ],
      "dimensions": {"dpi": 200, "height": 5139, "width": 2387}
    }
  ],
  "model": "mistral-ocr-latest",
  "usage_info": {"pages_processed": 1, "doc_size_bytes": 199841}
}

Base64 data of images are removed because converted images are already stored separately. This make JSON cleaner and smaller.

Limitations

• The script requires an active internet connection to access the Mistral API
• Processing large files may take time depending on your network speed
• The quality of text extraction depends on the clarity of the original documents
• API usage may be subject to rate limits or costs, depending on your Mistral account

License

MIT License

Acknowledgements

• This tool uses the Mistral OCR API