Text recognition (OCR)
Your scans become text — the analysis stays on your machine.
Your files never leave your deviceDrop a PDF or an image here
Scanned PDF or photo — JPG, PNG
How to recognize text in a scanned PDF (OCR), step by step
OCR turns a scan into usable text by loading a recognition engine directly in your browser: the document is never sent online.
Drop in a scanned PDF or a photo
Drag a PDF or a JPG/PNG image onto the drop area, or click to browse for it.
Choose the document’s language
Pick it under “Document language”: the site’s languages first, then more than a hundred others, whose model is downloaded on first use. Recognition is far less reliable if the chosen language does not match the actual text; “Second language” is for bilingual documents.
Narrow the page range, for a PDF
The field accepts a range like “1-3, 5, 8-10”; left blank, every page is analyzed.
Click “Start recognition”
The recognition engine (about 4 MB) downloads once from our servers, along with the language model; for a language that is not included, the model’s size is shown before you start. The analysis itself then runs entirely on your device.
Read the result and its confidence level
The recognized text appears with a word count and an average confidence score; a low score points to a scan that is probably skewed or poor quality.
Download the text or the searchable PDF
The .txt file holds the plain text; the searchable PDF keeps the original layout with an invisible text layer on top, which makes the document selectable and searchable by keyword.
In the searchable PDF, a scanned page keeps its original image and gets an invisible text layer that works for every script, Greek, Cyrillic, Arabic or Chinese included; a page that already held text is rebuilt as an image so that it does not carry two layers.
Frequently asked questions
Does VellumPDF’s OCR accept both PDFs and photos?
Yes, you can drop in a scanned PDF or a JPG or PNG image directly. Either way, the tool produces a searchable PDF whose text can be selected and copied.
What languages does the text recognition support?
More than a hundred languages. The site’s languages are included and served by VellumPDF; the others (Czech, Greek, Hebrew, Persian…) are picked from the same “Document language” list, and their model is downloaded once, on first use. “Second language” handles a bilingual document. Choosing the right language clearly improves the result.
Is my scanned document sent to an online artificial intelligence service?
No. The recognition engine runs in your browser. For the site’s languages, its files come from VellumPDF’s own domain; for any other language, only that language’s model is downloaded from raw.githubusercontent.com, the Tesseract project’s public repository, after a notice on screen. That download goes one way only: the document never leaves your device and no artificial intelligence service is called.
Is the recognition model downloaded every time I use the tool, and how large is it?
Each language has its own model. For the site’s languages, hosted by VellumPDF: 0.4 to 3 MB compressed depending on the language (about 700 KB for French, nearly 3 MB for English); the browser’s usual cache avoids downloading them again. For the other languages, between 0.4 and 12 MB depending on the language, with the size shown before you start; once downloaded, the model is kept by the browser and the second use downloads nothing.
How can I tell if the result is reliable, for instance on a skewed or poor-quality scan?
An average confidence percentage is shown alongside the result, calculated across every recognized word. Below 70 %, a warning flags the recognition as uncertain and the scan as possibly skewed or low quality — a cue to rescan rather than trust the text as-is.
What happens if I drop in a password-protected PDF?
Recognition is refused outright, with a message explaining that an encrypted PDF cannot be run through OCR. The protection needs to be removed from the document first, before dropping it in here.