I ran into the same problem a few months ago when I was building a payment verification feature. What really helped me was checking out this tool OCR studio
https://ocrstudio.ai/bank-card-scanner/. It’s surprisingly easy to plug into Python, Java, C#, or even C++ projects. I just fed the card images, and the OCR handled the text extraction really consistently. No need to train models myself, which saved me tons of time. I also liked that it could handle different orientations and blur levels without much tweaking.