OCR explained: getting the text out of a scan or photo
October 9, 2026 · 3 min read
You scan a contract, open the PDF, try to select a sentence to copy — and nothing happens, or the whole page highlights as one block. That's because a scanned document is a photograph of a page. It looks like text to you, but to the computer it's a grid of pixels. OCR, optical character recognition, is what turns that picture back into actual text.
Two kinds of PDF
| "Real" (digital) PDF | Scanned PDF | |
|---|---|---|
| How it was made | Exported from Word, an invoice system, a website | Photographed or scanned from paper |
| Can you select and copy text? | Yes | No, until it's OCR'd |
| Can you search for a word? | Yes | No, until it's OCR'd |
| File size | Small | Larger (it's images) |
A quick test: try to select a single word. If you can, the PDF already contains text. If you can't, it's a scan.
What OCR does
OCR looks at the image, finds the areas that contain text, and recognizes each character and word. Modern OCR — including the AI-based recognition CrunchFile uses — reads whole lines in context, which is why it copes with different fonts, uneven spacing and even neat handwriting far better than older systems that matched one letter at a time.
The result is the document's text as plain text: something you can copy into an email, paste into a spreadsheet, search with Ctrl+F, or edit.
What you can do with the text
- Copy instead of retyping: invoice numbers, totals, addresses, policy numbers.
- Fix a typo or update a detail in an old document that only exists on paper.
- Pull a table (a rate sheet, a parts list) into rows and columns.
- Summarize or translate a long document once its words are available.
- Redact sensitive details — account numbers, SSNs — before you share a copy.
In CrunchFile, the OCR / Extract Text tool pulls the text from an image or scanned PDF, and the other AI tools (summarize, translate, table extraction, redaction) work from that text.
Why OCR gets things wrong — and how to help it
OCR is only as good as the image it's reading. The usual culprits:
- Blur. Motion blur and missed focus are the biggest cause of errors. Hold still, tap to focus.
- Shadows and low contrast. Grey text in a shadowed corner can be misread or skipped. Clean the scan first: straighten it and use a black-and-white or shadow-removal filter, then run OCR.
- Tilt and curl. Text at an angle, or curving toward a book's spine, is harder to read. A straightened page reads better.
- Low resolution. Tiny print needs enough pixels. If you can't read the smallest text when you zoom in, OCR can't either.
- Messy handwriting, stamps over text, faded carbon copies. Expect to check and correct these by hand.
Always proofread numbers. OCR mistakes look plausible: a 5 read as an S, a 1 as an l, an 8 as a B. For anything that matters — amounts, account numbers, dates — compare the extracted text against the image before you rely on it.
A good routine for scanned paperwork
- Scan the document and let it straighten the pages.
- Clean it with B&W (text documents) or Color Enhance (documents with colored stamps).
- Run OCR to extract the text.
- Check the important numbers against the page.
- Use the text — copy it, edit it, or feed it to the summarize, translate or table tools.
The cleaner the scan, the less fixing you'll do afterwards.