Publié
What OCR Is, and When You Actually Need It for Your Documents
You forward yourself a scanned contract, open it, search for a name — and nothing happens. The page is a picture of text, not text. OCR is the bridge between the two. This is a plain explanation of what optical character recognition does, the single test for when a document actually needs it, and why the results are worth a second look.
You forward yourself a scanned contract, open it on your phone, and try to search for a name. Nothing happens. You try to copy a line, and nothing selects. The page looks like text, but the app treats it as one flat image — because that is exactly what it is.
The thing standing between a picture of words and words you can use has a name: OCR. Most documents you read will never need it. But knowing which ones do, and what to expect when you reach for it, saves a surprising amount of fumbling.
What OCR is
Optical character recognition turns an image of text — a scan, a photo, a screenshot — into machine-readable characters you can select, search, and copy. A scanner or camera captures a page as colored dots. To the device, those dots are a photograph; the letters carry no more meaning than the shadows around them. OCR reads the shapes and works out which characters they represent.
The process is less mysterious than it sounds. The software first cleans up the image — straightening a skewed page, dropping specks, separating ink from paper. Then it recognizes the characters; modern engines use neural networks that read whole lines at once instead of guessing one letter at a time. A final pass checks the work against a dictionary and fixes the obvious slips.
Here's the detail worth holding onto: good OCR doesn't throw your page away and retype it. It keeps the original image and lays a hidden layer of text underneath, so the document looks exactly the same while the words quietly become live.
When you actually need it
There is one test, and you can run it in a second. Try to select a line of text. If your finger or cursor highlights the words, the page already carries a text layer and needs nothing more. A PDF exported from software — a contract, an invoice, an ebook receipt — is "born digital" and arrives with its text already embedded. So do EPUBs, Word files, and Markdown.
You need OCR only when the page is image-only: a scan from an office copier, a phone photo of a book chapter, a faxed form, an old letter someone digitized years ago. They look like documents but behave like pictures. That is the moment OCR earns its place — when you want to search a two-hundred-page scanned PDF for a single clause, lift a quote out of a photographed page, or pull a figure from a report that only ever existed on paper.
It helps to be honest about the limits before you lean on it. Reported accuracy has ranged from roughly 81% to 99%, and even a one-percent character error can read as a five-percent word error once you count whole words. Clean print scans beautifully. Glare, skew, faint ink, unusual fonts, and especially handwriting all pull the numbers down. Treat the result as a strong draft you glance over, not a transcript you trust blindly.
Where you'll see it
OCR is mostly invisible, and it is already all around you. It makes scanned book archives searchable, reads the account numbers off bank checks, scans passports at airport gates, captures license plates, and turns a business card into a contact. None of it announces itself as "OCR" — it is just the quiet step that makes paper behave like data.
It lives on your phone, too. Point a camera at a street sign in another language, and the translation that appears rests on text recognition. Apple's Live Text lets you copy a phone number straight out of a photo, look up a word, or share a paragraph you only have as a picture. What was once a specialist's tool has settled into the devices you already carry.
How Margina fits in
Margina is a local-first reading app for the documents you send yourself — PDFs, EPUBs, Word files, the chapters and contracts that pile up in your own inbox. Those self-forwarded scans are exactly the files most likely to need OCR, which is why it's worth understanding before you wonder why a page won't let you search it.
There is a privacy point here, and it is a simple one. On-device text recognition reads the page where it sits rather than uploading it somewhere to be processed — the same posture a local-first reader takes with everything you open. Your documents stay yours.
If you want to keep reading along these lines, there are more guides on reading documents. And if you'd rather just try it, Margina is on the App Store and Google Play.
Frequently Asked Questions
How can I tell if a PDF needs OCR?
Try to select or highlight a line of text. If words highlight, the page already has a text layer and you don't need OCR. If nothing selects — the whole page acts like a picture — it's an image-only scan, and OCR is what would make it searchable.
Is OCR always accurate?
No. Reported character accuracy has ranged from about 81% to 99% depending on the document, and even a one-percent character error can mean roughly a five-percent word error rate. Clean, well-lit print scans best; glare, skew, faint ink, unusual fonts, and handwriting all lower accuracy.
Does OCR change my original document?
Good OCR doesn't replace the page. It adds an invisible text layer beneath the original image, so the document still looks exactly the same while the words become selectable and searchable.
Can OCR read handwriting?
Sometimes. Modern engines can detect handwritten text alongside print, but cursive and messy writing remain the hardest cases, and the results are worth checking before you rely on them.
In short
OCR is the bridge between a picture of text and text you can actually use. You reach for it only when a document won't let you select its words — most won't need it, but a scan will. And as more of our reading drifts onto the phone, that bridge keeps getting shorter, quieter, and easier to forget is even there.