What is OCR, and why do scanned PDFs need it?

Last reviewed: September 23, 2026 · Markdown version

OCR — optical character recognition — is software that turns a picture of text into text a computer can actually use. It looks at the image, recognizes letters and words, and stores them in machine-readable form. That's the whole trick, and it's the difference between a page you can only look at and a page you can search, select, and copy.

Why scanned PDFs need it

A scanned page is a photograph. To your eyes it's a contract or a statement; to a computer it's pixels, with no words in the file at all. Search reads a file's stored text, finds none, and reports nothing — the classic "the word is right there and search can't find it". OCR is the bridge: after it runs, the recognized words are stored alongside the page image, and search has something to read.

The question nobody asks: where does the OCR run?

This is a privacy question wearing a technical costume.

Cloud OCR uploads your page images to a company's servers and sends text back. It works — and your bank statement, contract, or medical record has now been on infrastructure you can't see or audit.
On-device OCR runs on your own phone, and the images stay there. Both phone platforms document a local route. Apple says of its Vision framework that "all of Vision's processing happens on the user's device to enhance performance and user privacy" (Recognizing Text in Images, read 23 September 2026). Google offers ML Kit's text recognition in a bundled form, where the "model is statically linked to your app at build time" and is "available immediately" (Google for Developers: ML Kit text recognition, read 23 September 2026). So for personal documents, uploading is a choice, not a necessity.

What OCR honestly can't do

OCR is recognition, and the vendors say plainly that input quality decides the result. Google's guidance asks for text "represented by sufficient pixel data" — ideally "at least 16x16 pixels" per character — and warns: "Poor image focus can affect text recognition accuracy." (Same ML Kit page.) Apple offers an optional language-correction phase in Vision "to minimize the potential for misreadings" (same Vision article), which is an admission that misreadings happen. Neither page, as read today, says how either engine copes with handwriting; treat a handwritten page as unsearchable unless you have tested it.

A misread has a sneaky consequence for search — if the scan says "invoice" but OCR stored "lnvoice", your search for the right spelling misses it. A trustworthy tool therefore does two things: labels which text came from OCR, and tells you which pages it couldn't make searchable — so a silent failure never masquerades as an empty document.

Where DocFind fits

DocFind takes the on-device route: it reads scanned pages on your phone — Apple's Vision on iPhone, Google's ML Kit bundled into the app on Android — and folds the recognized words into a searchable index across your PDF library. Reading scanned pages works with no internet connection, and DocFind never uploads your documents. Matches found through OCR are labelled as such, and pages that couldn't be made searchable are reported — the honest accounting this page has been describing. It is built for printed text, not handwriting.