What Is OCR (Optical Character Recognition) in AI?
OCR is how AI reads text that only exists as pixels — a scanned document, a photo of a sign — turning an image of text into text a model can actually process.
Reading text out of an image
Optical character recognition (OCR) detects and extracts text from an image — a scanned document, a photo of a street sign, a screenshot — converting what's visually pixels into actual machine-readable text a system can search, edit, or feed into a language model.
Why this is different from a vision-language model just "reading" an image
A vision-language model can describe an image in general, but dedicated OCR is specifically optimized for the narrower, more precise task of getting every character right — important for use cases like extracting an exact total from a receipt, where "roughly right" isn't good enough.
What makes OCR harder in practice
Real-world OCR has to handle unusual fonts, poor image quality, skewed angles, and handwriting, which is meaningfully harder than recognizing clean, printed text. Modern OCR systems increasingly use the same underlying architectures as other vision AI to handle this messier, real-world variation.
Frequently Asked Questions
What is OCR?
Optical character recognition — AI technology that detects and extracts text from an image, converting pixels into actual machine-readable text a system can search, edit, or process further.
Why is OCR treated as a separate capability from general image understanding?
OCR is specifically optimized for getting every character exactly right, which matters for precise use cases like extracting a total from a receipt — a narrower, more exacting task than a model just generally describing an image.