What is optical character recognition (OCR)?
Optical character recognition is the conversion of text in an image — a scan, a photograph, a fax — into machine-readable characters. Modern OCR combines computer vision with language models, which lets it resolve ambiguous characters from surrounding context rather than reading each glyph in isolation.
OCR is the first stage of nearly every document pipeline, and its accuracy sets a ceiling on everything downstream. If a page is read at 92% character accuracy, no amount of clever classification or extraction afterwards recovers the missing 8%.
The difficulty in production is rarely clean, flat, high-contrast pages — those were solved decades ago. It is phone photographs taken at an angle, carbon copies, handwriting in the margins, stamps overlapping printed text, and decades-old microfiche. Systems that report high benchmark accuracy often fall over on exactly this material, which is why accuracy should be measured on your own documents before committing to an approach.
Where this shows up in our work
Related terms
Got an idea? Let's make it real.
Tell us about your problem. We'll come back within one business day with a take, a rough plan, and a call invite.