Hello Johan Ögren ,
Thank you for reaching out on Microsoft Q&A.
To improve the accuracy of Azure AI OCR for reading punched numbers and letters on plates, consider these effective steps: Firstly, while you can't directly train the Azure OCR service, you can create a custom OCR model using Azure Custom Vision by providing high-quality images with correct text labels, covering various fonts and lighting conditions to enhance recognition capabilities. More information is available here. Additionally, preprocessing images can significantly boost accuracy through techniques such as enhancing contrast, reducing noise, and ensuring proper cropping and resizing. Azure Cognitive Services offers Form Recognizer, which is useful for structured data like license plates; you can explore this option here. By combining these methods, you can significantly enhance the OCR accuracy for punched numbers and letters on plates.
If you found this solution helpful, consider accepting it.