Image to Text (OCR) — English, Urdu, Arabic
Extract text from a photo, screenshot or scanned page with on-device OCR in English, Urdu, Arabic, Hindi and more — copy or download the text, nothing uploaded.
In short
On-device OCR with Tesseract (WebAssembly) extracts printed text from photos, screenshots and scans in English, Urdu, Arabic, Hindi and other languages without uploading the image; text should be at least about 25 px tall, and numbers, names and IDs always need proof-reading.
Who this is for: Students, office staff and freelancers who need the text out of a photo, screenshot or scan without retyping it.
How on-device OCR works and how to get good results
This page runs Tesseract — the open-source OCR engine behind many commercial tools — compiled to WebAssembly so it executes inside your browser. The image is analysed locally; only the language model is downloaded (and cached) the first time you use each language.
- Resolution: text should be at least ~25 px tall in the image. Photograph documents straight-on in good light; crop to the text block with the crop tool if the page has a busy background.
- Languages: English is the most accurate. Urdu uses the urd model, which handles printed Nastaliq and Naskh reasonably but not handwriting; Arabic (ara) is strong on printed Naskh. Mixed documents (an Urdu letter with English names) use the combined models.
- Numbers and IDs: OCR confuses 0/O, 1/l/I and 5/S — always proof-read CNIC, Iqama, IBAN and phone numbers.
- PDFs: convert the page to an image with the PDF-to-JPG tool first, at 200–300 dpi.
Assumptions
- Tesseract.js v5 worker created for the selected language(s); language data fetched from the public tessdata CDN and cached.
- Confidence and word count come from the engine's result; text direction set automatically for Arabic-script output.
- The image is processed in the browser only.
Worked example
A 12 MP phone photo, cropped to the bill, English → about 180 words extracted in 5–10 seconds on a laptop (longer on a phone), confidence ~90%; account numbers worth checking by eye.
Frequently asked questions
Is my image uploaded to a server?
No. The OCR engine runs in your browser via WebAssembly; the only network request is for the language file from a public CDN.
Why is Urdu accuracy lower than English?
Nastaliq script joins and stacks letters in ways that are hard to segment; the open-source Urdu model is trained mainly on printed text. Use a high-resolution, straight image and expect to correct some words.
Can it read handwriting?
Only neat block capitals with luck. Cursive handwriting is not supported by Tesseract.
Why does the first run take long?
The language model (10–20 MB) and the engine are downloaded once; later runs reuse the browser cache and take seconds.
How do I copy text from an image?
Choose the image and language, click Extract text, then Copy — OCR runs in your browser.
Can it read Urdu?
Yes, with the Urdu model, best on clear printed text; expect some corrections.
Related calculators
Found an error? Report it — tell us the inputs, the result you got, what you expected and the rule or source. Confirmed corrections are published within a week.
Ready to build your next digital solution?
Send the brief and you get a scoped response with approach, milestones and a fixed price range. Not a sales call.
A 30-minute call with the engineer who would run your project
Within 2 business daysA written scope: approach, module breakdown, milestones, fixed price range
Within 5 business daysYou decide. No retainer, no obligation, and the scope document is yours to keep either way
Your timelineSend your brief
Three fields to start. Add project details if you have them.