What text recognition actually does
What optical character recognition is, what it cannot do, and why the scan matters more than the software.
OCR stands for optical character recognition. It is the process of looking at a picture of writing and working out what the writing says.
The problem it solves
A scanner produces an image. To the machine, a scanned letter is a grid of light and dark dots that happens to look like words to you — there is no “A” in the file, only a pattern of pixels that a human reads as an A.
Recognition closes that gap. It finds the regions of the image that are text, works out where the lines and words and characters are, decides which character each shape is, and writes the answer out as actual text. That text is then added to the PDF underneath the picture, so the page looks exactly as it did and the document now contains words.
What it makes possible
Everything that depends on a document containing words rather than looking like it does:
- Searching the document, and finding it with your Mac's own search
- Selecting and copying a passage instead of retyping it
- Editing the text in place
- Highlighting and underlining, which need something to select
- Having it read aloud
- Using it with a screen reader at all
- Converting it to Word, Excel or plain text with anything in it
What decides whether it works well
| Factor | Effect |
|---|---|
| Resolution | Below about 200dpi, characters lose the detail that distinguishes them. 300dpi is the usual sensible minimum for text. |
| Straightness | A page scanned at an angle is harder to segment into lines. Straightening it first helps more than any other single thing. |
| Contrast and lighting | A photograph with a shadow across the page, or grey text on grey paper, is harder than clean black on white. |
| Typeface | Ordinary text faces recognise very well. Decorative, condensed and very small type recognise less well. |
| Language and characters | Accented characters, non-Latin scripts and mathematical notation are harder than plain text. |
| What else is on the page | Handwriting over print, stamps across text and heavy marks all interfere. |
No recognition is perfect. A well-scanned page of ordinary text comes out very accurate indeed; a fax of a fax does not. Because the original picture is left untouched, a misreading affects search and copying rather than how the document looks — but it does mean you should check anything you are going to act on, particularly numbers.
What recognition does not do
- It does not understand the document. It converts shapes to characters. It does not know which line is a heading, which number is a total, or what the document is about.
- It does not fix a bad scan. Detail that was not captured cannot be recovered, however good the software.
- It does not rebuild the layout. Tables in particular recognise as text in positions rather than as a structured table, which is why converting a recognised scan to a spreadsheet gives a rougher result than converting a born-digital one.
- It does not read handwriting reliably. Printed text is a solved problem; handwriting is not.
Where it runs, and why that matters
Recognition is real computation, and for years that meant sending the document to a server with the horsepower to do it. Modern Macs do not need to. Verso runs recognition on your machine, which is the difference between a feature you can use on a medical record and one you cannot.
Frequently asked questions
What does OCR stand for?
Optical character recognition — looking at a picture of writing and working out what it says.
Does OCR change how my document looks?
No. The picture of the page is left exactly as it was; the recognised text is added underneath it. The document looks identical and behaves completely differently.
Why is my OCR result inaccurate?
Almost always the scan rather than the software: low resolution, a crooked page, poor contrast, an unusual typeface, or handwriting over print. Straightening and rescanning at 300dpi fixes most of it.
Can OCR read handwriting?
Not reliably. Printed text recognises very well; handwriting remains a much harder problem.
Does OCR need an internet connection?
Not in Verso. Recognition runs on your Mac, which is why it can be used on documents that must not leave the machine.
Will OCR turn my scanned table into a spreadsheet?
Partly. Recognition produces text in positions rather than a structured table, so converting a recognised scan to a spreadsheet gives a rougher result than converting a document that was born digital.