Optical character recognition reads the words inside pictures of text: scanned contracts, photographed receipts, screenshots, pages of a scanned PDF. This tool runs that recognition entirely in your browser — PDFs are rendered page by page and each page is read by a text-recognition engine, so the file never leaves your device. Copy the result or download it as a plain text file.
How it works
Image files go straight to the recogniser. PDF pages cannot be read directly, so each page is first rendered as a high-resolution image (twice normal size) and then recognised one page at a time, with the results joined in page order. The engine is a trained neural network that examines the shapes of letters in context rather than matching templates, which is why it handles ordinary fonts well and decorative ones badly. The language data it needs downloads once, on first use, and is then cached by your browser. PDFs longer than twenty pages are processed for the first twenty only — beyond that, split the file first and run each part separately.
A worked example
Photograph a printed electricity bill in decent light, making sure the page fills the frame and the text is sharp. Upload the photo, keep English selected, and start extraction: the tool reports progress as it reads, then returns the bill's lines as plain text — account number, amounts, due date — ready to copy into a spreadsheet. Rendered PDF pages behave the same way, one after another. Before extracting, it can help to preview exactly which pages hold text using our PDF to image converter, and to enlarge a small or low-resolution image first with the image resizer.
What limits accuracy
Recognition is only as good as the pixels. Small, blurry, skewed or shadowed text degrades quickly — anything under about 150 dots per inch is guesswork, and phone photos taken at an angle should be retaken straight-on. Handwriting is not reliably readable; joined-up script fails outright. Complex layouts (tables, two columns, text over images) may come out in the wrong reading order, and decorative or unusual fonts confuse the engine. Finally, results are only as correct as the language selected: run German text with the English data and accuracy collapses, so match the language to the document.
Privacy, honestly stated
Your file is processed locally and never uploaded anywhere — that is the point of running the engine in the browser. Two honest caveats. First, the language data file downloads from a public content network on first use, which is a normal file download, not your document going anywhere. Second, extraction happens on your device, so a very large PDF will use noticeable memory and take a while; nothing about that leaves your machine either.