DuckConvert Mascot
DuckConvert
Blog

OCR PDF

Make a scanned PDF searchable, in your browser. The page images stay exactly as they are — the text is added invisibly on top.

Make a scanned PDF searchable

Reads the words in each page image and lays them over the page as invisible, selectable text. The scan itself is left exactly as it is.

What a searchable scan actually is

A scanned PDF is a stack of photographs. Your reader shows you what looks like a document, but as far as the file is concerned there are no words in it at all — which is why searching finds nothing and selecting a sentence gives you a rectangle of image.

The fix is a second layer. The page image stays where it is, and the recognised words are drawn over it in a mode the PDF specification provides precisely for this: the glyphs take part in selection, search and copy, and paint nothing at all. Your viewer highlights text that is invisible, sitting exactly on top of the picture of that text.

Why the positioning is the hard part

The recogniser reports each word as a box measured in the pixels of the image it was shown. A PDF page counts from the bottom-left in points rather than the top-left in pixels, has a scale of its own, and may carry an instruction telling the viewer to turn it a quarter turn before drawing anything. Every one of those has to be undone to put a word back where it came from.

Get it wrong and the result is worse than doing nothing: the document looks perfect, and selecting a line hands you words from somewhere else on the page. That is why each word is stretched to the exact box it was found in rather than guessed at a font size.

What it will not do

It will not improve a bad scan. Recognition quality is set by what the scanner captured — faint text, heavy skew, a photo taken at an angle, or handwriting all read poorly, and no amount of processing afterwards recovers detail that was never there. If a document matters, rescanning it straight at 300 DPI is worth more than any tool on this page.

OCR PDF FAQ

Is my document uploaded to be read?

No. Recognition runs on your own device: the text engine is WebAssembly your browser downloads once, and the pages never leave the tab. That is unusual for this particular tool — OCR is expensive, so nearly every other service on the web does it on their servers, which means the contract, medical record or bank statement you are trying to make searchable gets uploaded to a stranger first.

Why does it take so long?

Because it is genuinely reading every page. Recognition is a few seconds per page on a normal laptop, so a twenty-page scan takes a couple of minutes. The progress bar names the page it is on so you can see it moving. A server-based tool feels faster because it is spending someone else's CPU, not because there is less work.

Does it change how my document looks?

Not at all. The original page images are kept exactly as they are and the recognised words are added on top as invisible text — the same trick every searchable scan uses. Nothing is re-rendered, re-compressed or replaced, so the file looks identical and is now searchable.

My PDF already has selectable text. Will this help?

Probably not, and you likely do not need it. This is for image-only PDFs, the kind a scanner or a phone camera produces. If you can already select a sentence in your reader, the text layer is there. Running this on such a file would add a second, redundant layer read off the rendered page.

How accurate is it, and what if it gets a word wrong?

It depends almost entirely on the scan. Clean, straight, 300 DPI text is read very well; a faint fax, a skewed photo, or handwriting will not be. The recognised text is invisible, so a mistake never shows on the page — it just means that word cannot be found by searching. Nothing visible is ever altered by a misread.

Which languages does it handle?

English at the moment. Each additional language needs its own trained data file to be mirrored and shipped, so more will be added deliberately rather than all at once.