OCR tools

Recognition that never leaves your device

Optical character recognition looks at a picture and works out which shapes are letters. Every mainstream OCR service does that on a server, which means uploading the image first — and the images people run through OCR are overwhelmingly documents. Passports, driving licences, payslips, tax paperwork, medical letters, and bank statements, because those arrive as scans and photographs and the text is locked inside a picture.

These tools run Tesseract compiled to WebAssembly, in your browser tab. The engine and its English language data are around 6 MB and download once, then stay cached. Recognition is real computation on your device, so it is slower than a server would be — a page takes a couple of seconds on a laptop and longer on a phone. That is the honest trade for the image never being transmitted.

Getting good results

Accuracy is almost entirely an input problem, and it is one you control. A flat, well-lit, straight-on photograph of printed text at decent resolution comes out close to perfect. Shadows across the page, a camera held at an angle so the lines converge, and low resolution are what wreck it. Two minutes spent retaking the photograph beats any amount of correcting afterwards.

Handwriting is the hard limit. Tesseract is trained on printed type, so neat block capitals are unreliable and cursive is effectively out of reach. And before reaching for OCR at all, check whether you need it: if you can select text in your PDF, PDF to Text reads what is actually stored and is both faster and exact. OCR is only for when the words exist as pixels.

Everything in ocr tools

• Image to Text (OCR) — Read the words out of a photo or screenshot.
• OCR PDF — Read text out of a scanned PDF.
• Image to Excel — Photographed table into a spreadsheet.
• Image to CSV — Photographed table as plain CSV.