Could you handle that by taking a screenshot, allowing the user to crop the area with the actual recipe text, then using OCR?