How to get a table out of a scanned PDF
A scan has no text in it to copy, so the usual advice does not apply. Here is what actually works, what it costs, and when to use something else.
Last checked 2026-09-21.
First, find out which kind of PDF you have
There are two sorts of PDF that look identical on screen. One holds real text; the other holds a photograph of text. Everything about getting data out depends on which you have.
Open it in any reader and try to select a word with your mouse. If you can highlight it, there is a text layer and most converters will work. If your cursor draws a rectangle over the page instead, it is a scan, and a converter that only reads text will hand you an empty file or an error.
A file produced by a scanner, a phone camera, or a fax is a scan. So, often, is a bank statement downloaded as a PDF image, or any document that has been printed and re-scanned at some point in its life.
What OCR does, and what it costs you
Optical character recognition looks at the picture and works out which shapes are which letters. It gives back words along with where each one sat on the page, which is the same raw material a text PDF provides, so the table can then be rebuilt the usual way.
Three costs are worth knowing before you start. It is slower: a second or two per page rather than instantly. It is not perfect: expect the occasional 8 read as a B, and more mistakes on faint or skewed scans. And most services do it on their servers, which means uploading the document.
Doing it without uploading the document
For a menu or a leaflet that hardly matters. For a bank statement, a medical form, a contract, or anything with someone's name and account number on it, handing the file to a company you have never heard of is a real cost that most people never price in.
It is avoidable. OCR can run inside your browser: the recognition engine is compiled to WebAssembly, downloaded once, and then everything happens on your own machine. It is slower than a server would be, and for a document you would rather not hand over, that is usually the right trade.
That is how our PDF to Excel tool handles scans. It notices there is no text layer, offers to read the pages with OCR, downloads the engine the first time, and does the rest locally. You can watch your browser's network tab while it runs: the document is not sent anywhere.
Getting the best result from a scan
OCR accuracy is mostly decided before you start, by the quality of the scan:
- 300 dpi is the sweet spot. Below 200 the error rate climbs steeply. Above 400 you mostly get a bigger file.
- Straight beats crooked. A page scanned at a slight angle confuses line detection more than it confuses letter recognition, so rows start merging.
- Flat beats curved. A photograph of a page in a bound book curves near the spine, and the text there reads worst.
- Grayscale or black and white usually beats color for text, and is much smaller.
- If you have any way of getting the original digital file instead of a scan of it, that beats every scanning tip on this list.
When to use something else
We would rather tell you this than have you waste an hour:
- A page of dense prose rather than a table. OCR will read it, but you want a text extractor, not a spreadsheet converter.
- Hundreds of pages, accuracy critical, budget available. Adobe's OCR is better than anything that runs in a browser, and for a long legal or financial job the subscription is cheaper than checking the output by hand.
- Handwriting. General-purpose OCR reads printed text. Handwriting needs a different class of tool and still struggles.
- A language in a different script. Our tool reads English. Other scripts need the matching trained data.
If the document is ordinary, printed, and yours, browser-side OCR will very likely do the job and nobody else will ever see the file.
Questions people ask
How accurate is OCR on a scanned table?
On a clean 300 dpi scan of printed text, high enough that the mistakes are occasional rather than constant, and our tool reports the confidence it achieved. On a faint, skewed, or low-resolution scan it degrades quickly. Always check any figure you are going to rely on.
Does OCR work on a photo of a document?
Often, yes, if the page fills the frame, is reasonably flat, and is evenly lit. A photo taken at an angle in poor light is the hardest case there is.
Why does browser OCR need a download the first time?
The recognition engine and its trained data are a few megabytes, and they have to be on your machine for the work to happen there. It downloads once and your browser caches it.
Try it
The OCR runs on your own machine, so a scanned statement or contract is never uploaded to anyone.