What is OCR and what does it do to a scanned PDF?

A form at the clinic asks you to "upload an OCR-readable copy", and your scanner app has a switch labeled OCR, but nobody says what it does. What is OCR, and does your lease, receipt or school form need it? This guide explains the term in plain words, what an OCR PDF holds that a plain scan does not, how accurate the software is, and how to get it on an iPhone.

What is OCR, and what does OCR stand for?

OCR stands for optical character recognition. Wikipedia defines it as "the electronic or mechanical conversion of images of typed, handwritten or printed text into machine-encoded text" (Optical character recognition, Wikipedia). In plain words, OCR looks at a picture of a page and writes down the letters it sees.

That matters because a scan is a photo. Your eyes read "Rent due on the 1st", but the phone sees colored dots, so search, copy and screen readers have nothing to work with.

A photo of text vs machine-readable text

A photo of text is pixels; machine-readable text is a string of letters that a computer can search, sort and paste. OCR is the bridge between the two. After OCR, the word "deposit" on page 2 of a lease is a word the computer knows, and a search can find it.

In a PDF, OCR usually adds a hidden layer. Wikipedia describes the result as "an annotated PDF that includes both the original image of the page and a searchable textual representation." The page looks the same as before; the text sits invisibly under the image, lined up with the words you see.

How OCR works inside the software

OCR software works in three stages: it cleans up the image, finds where the text is, and then decides which character each shape is. Wikipedia lists the clean-up steps, and the most common ones are easy to picture.

• Straightening (deskew): a page photographed at a slight angle is turned "a few degrees clockwise or counterclockwise in order to make lines of text perfectly horizontal or vertical."
• Black and white (binarization): the image is converted to two colors, "a simple way of separating the text … from the background."
• Layout analysis: columns, paragraphs, tables and captions are marked as separate blocks, so a two-column letter is read in the right order.
• Line and word detection: each block is cut into lines, then words, then single characters.

Recognition comes next. Older systems compared each character "to a stored glyph on a pixel-by-pixel basis"; newer ones break letters into features such as lines, loops and crossings, and many now use neural networks. Apple's documentation for its Vision framework says its accurate path "uses a neural network to find text in terms of strings and lines," which is "much more in line with how humans read text" (Recognizing text in images, Apple Developer).

Why a dictionary helps, and when it hurts

A dictionary helps OCR fix its own guesses. Wikipedia notes that "OCR accuracy can be increased if the output is constrained by a lexicon," a list of allowed words, so a misread such as "rnoney" can be pulled back to "money". Apple's Vision offers a similar optional language-correction step "to minimize the potential for misreadings."

The same trick can hurt with words outside the dictionary. Surnames, street names, product codes and account numbers are the classic victims, so those are the parts of a page to check by eye.

How to OCR a paper document on iPhone

The fastest way to OCR paper on an iPhone is to scan it with an app that adds the text layer as it saves. In MRS, that happens to every page by default; the steps below use a 3-page lease as the example.

1. Open the scanner. In MRS, tap "Scan document". For a till slip, tap "Scan receipt" instead, which also reads the shop, date, total, tax and item lines.
2. Capture every page. Hold the iPhone over the first page and let Apple's document camera find the edges and take the shot, then turn to the next page in the same session. All pages land in one PDF.
3. Let MRS add the text layer. MRS runs Apple Vision text recognition in accurate mode with automatic language detection on each page, on the iPhone. Nothing is uploaded and no account is needed.
4. Look at the page. The scan looks the same as before, because the recognized text is invisible and sits under the image.
5. Select or copy text. Open the PDF in Files or any PDF app, press and hold a word and drag the handles to select a line. Copy it and paste it into Notes to see how clean the result is.
6. Find the document by a word inside. Type a word from the body of the lease, such as the landlord's surname, into the MRS library search; it matches title and text, upper or lower case. To jump to that word on the page, open the PDF in Files, Books or Safari, as the PDF search guide shows.

The part people get wrong: OCR happens when you scan

MRS reads the text when a page is scanned or imported as a photo. PDFs you import from the Files app keep their bytes unchanged: if they already contain text, the library search reads it, and if they are image-only, MRS does not run OCR on them.

Documents scanned in MRS before text recognition arrived have no text layer, so the library finds them by title only; MRS does not reprocess old files in bulk. For an old image-only PDF, the searchable PDF guide covers the desktop and online routes.

OCR vs Live Text vs a plain scan vs handwriting recognition

Four terms come up when you read about scanning, and each gives you something different. The rows below stick to what the cited sources and the MRS feature list state.

Live Text is the right pick for a single line from a photo, and the Live Text guide has the taps. For paper you file and search for months, an OCR PDF keeps the text with the document.

Where OCR runs: on your phone or in the cloud

OCR runs in one of two places, and the difference is where your page goes. Wikipedia notes that OCR is "generally an offline process" and that "there are cloud based services which provide an online OCR API service"; a cloud service needs the image of your page on its server to read it.

On-device OCR keeps the page on the phone. Apple states that with its Vision text recognition "all of Vision's processing happens on the user's device to enhance performance and user privacy." MRS uses that framework, so a bank letter, a medical form or a passport page is read on the iPhone and is not uploaded.

How accurate is OCR?

OCR accuracy is high on clean print and never 100%. Wikipedia states that typewritten Latin-script text "is still not 100% accurate even where clear imaging is available," and it cites a study of 19th- and early 20th-century newspaper pages in which character accuracy of commercial OCR "varied from 81% to 99%."

How you measure it changes the number. The same Wikipedia article gives an example: a 1% character error rate (99% accuracy) can become a word error rate of 5% or worse, because one wrong letter spoils a whole word.

The page you feed in decides most of the result:

• Light: even daylight beats a lamp that throws your phone's shadow across the sheet.
• Flat page: a curled page bends the lines, and letters near the fold come out wrong.
• Size and sharpness: fill the frame so small print gets enough pixels per letter; hold still until the scan is sharp.
• Fonts: plain printed fonts read best; decorative scripts and very thin type read worse.
• Paper condition: faded thermal receipts, stamps over text and low-contrast copies lose letters.

The scanning guide covers light, angle and color modes in more detail.

A rule for receipts and handwriting

Receipts and handwriting are the two hardest everyday cases. Thermal receipts fade, so scan a receipt the day you get it, in good light; the receipt scanning guide shows the full routine and where MRS saves the text file (On My iPhone › MRS › Receipts).

Handwriting is harder still: Wikipedia states that cursive "has recognition rates even lower than that of hand-printed text." For handwritten notes, put the words you will search for in the PDF title, because the text layer may miss them.

What an OCR PDF lets you do, and its limits

An OCR PDF turns a stack of scans into documents you can work with. Four uses cover most of what readers ask about.

• Search: find the one lease, warranty or school letter that mentions a word, out of dozens of files.
• Copy: paste an account number, an address or a paragraph into an email or a form without retyping.
• Receipts to data: turn the shop, date and total on a receipt into text for an expense list.
• Reading aloud: let a screen reader speak the page to someone who cannot read the print.

OCR also has limits that catch people out, and most of them have a two-minute fix.

• Mistake: expecting an editable document. OCR makes a scan searchable and copyable; the page stays an image. To add your own text or a signature on top, use a PDF editor.
• Mistake: trusting numbers without a check. A 5 read as an S or an 8 as a B happens on blurry pages. Compare totals, dates and IBANs with the image before you copy them into a form.
• Mistake: scanning at an angle in a shadow. Skewed, dark pages give the recognizer broken letters. Let the scanner find the edges and move to even light.
• Mistake: naming files "Scan 12". Search by words inside is a safety net; a clear title such as "Lease 2026 Oak Street" still finds the file fastest.

Screen readers and accessibility

Accessibility is the use with the highest stakes. The W3C, which writes the web accessibility guidelines, calls a scanned document "inherently inaccessible because the content of the document is images, not searchable text," and its PDF technique says "OCR converts images of words and characters to actual text" (W3C PDF7). A blind relative, a student with dyslexia or a colleague using VoiceOver can only hear a page that has a text layer.

Your OCR checklist and next steps

Run through these before you file a scan away.

• Scanned with OCR: the page came from a scan or a photo, so it has a text layer.
• Tested: you selected a word in the body text and a search found it.
• Checked numbers: totals, dates and account numbers match the image.
• Named well: the title holds the words you will type later.
• Kept private: sensitive pages went through on-device OCR, not an upload.

MRS is free to download for iPhone with iOS 16 or later, in 33 languages, and reads every new scan on the device. The book scanning guide covers long documents, and you can download MRS to make your next scan searchable from the first page.