OCR stands for Optical Character Recognition. It's the technology that looks at a picture of text — a photo, a screenshot, or a scanned page — and figures out the actual letters and words, giving you real text you can copy, search and edit.
Without OCR, a scanned document is just an image: you can see the words but you can't select or search them. OCR adds a hidden (or extracted) text layer so the content becomes usable.
Common uses include digitising printed documents, making scanned PDFs searchable, and pulling text out of a photo.
You can OCR a scan with OCR PDF or extract text from a picture with Image to Text.
How OCR works
Optical Character Recognition turns a picture of text into actual text. The engine cleans up the image, finds the regions containing writing, works out where lines and characters begin and end, classifies each character shape, and finally checks the result against a language model so context can resolve visually ambiguous cases. That last stage is why picking the right language matters: recognising Romanian with an English model produces markedly worse output, because the dictionary used to settle ambiguity is the wrong one.
What a searchable PDF actually is
Running OCR over a scan does not replace it with typed text. It adds an invisible text layer positioned behind the page image, so the document looks exactly as it did while becoming searchable, selectable and readable by screen readers. Because the original image stays on top, a recognition error never changes what a person sees — it only affects what a search finds. This is also why OCR is safe to run on documents that must look untouched.
What it handles well and badly
Clean, high-contrast, roughly horizontal printed text is recognised accurately: scanned documents, printed pages, screenshots. Accuracy falls sharply on faint photocopies, skewed or curved pages, decorative fonts, low-resolution photographs, and text over busy backgrounds. Handwriting is largely out of reach for engines built for print — neat block capitals sometimes come through, cursive generally does not.
OCR before compression, not after
Order matters. OCR reads the page image, so it works best on the sharpest version available. Compressing first discards exactly the fine detail the recogniser depends on, and character accuracy drops noticeably. Run OCR on the original, then compress the result: the text layer is tiny and adds almost nothing to the file size.
Why doing it locally matters
OCR is routinely run over passports, contracts, invoices and medical letters — precisely the documents you would least like to hand to someone else's server. Running recognition in your own browser means the file never leaves your device. It is slower than a datacentre, but nothing is transmitted, retained or logged.