optical character recognition (OCR)
/ OP-tih-kul KAIR-ik-ter rek-ug-NISH-un (O-C-R) /
Optical character recognition turns pictures of text into actual text — the kind a computer can search, copy, and edit. To a camera, a printed page is just a pattern of dark and light; OCR reads that pattern and recovers the letters and words. It is what lets you photograph a receipt and have the numbers become editable, snap a foreign menu and translate it, or take a century-old scanned book and make every word searchable. It bridges the world of ink and the world of characters.
A typical system works in stages: find the regions that contain text, straighten and clean them up, segment them into lines and characters, recognize each shape as a letter or symbol, and often a language model on top fixes likely mistakes ("the" not "tlie"). Clean, printed, high-contrast text — a typed document, a license plate — is now solved extremely well, often near-perfectly. That reliability is why OCR quietly runs inside translation apps, document scanners, mail sorting, and accessibility tools that read text aloud.
The difficulty climbs steeply as text gets messier. Handwriting — especially cursive or sloppy — is dramatically harder than print. Curved text on a bottle, faded or stained documents, dense tables, unusual fonts, low light, and motion blur all drive errors up. And recognizing the characters is only half the battle: reconstructing layout and reading order in a complex magazine page or a multi-column form is its own hard problem. Languages with thousands of characters or no spaces between words add further challenge. Treat any OCR output as a draft to verify, not a guaranteed transcript.
Photograph a printed business card and OCR returns the name, phone, and email as editable text, ready to save as a contact — usually flawless. Photograph a handwritten grocery list and it might read "milk" fine but turn "eggs" into "e995" and miss a word that ran off the torn edge.
Near-perfect on clean print, error-prone on handwriting and clutter — the gap is the whole story.
OCR on clean printed text is largely a solved problem; OCR on handwriting, curved or degraded text, and complex layouts is not. Recognizing the characters is also separate from reconstructing reading order — a perfectly read multi-column page can still come out scrambled.