What OCR does, and where it struggles with Urdu
OCR (optical character recognition) turns a picture of text into real text that you can edit, search and copy. It is the way to rescue Urdu that only exists as an image: a screenshot shared on WhatsApp, a photo of a notice, a scanned application form, or a PDF made by a scanner.
This tool uses Tesseract, the free, open-source OCR engine originally developed at HP and later by Google, together with its official Urdu model. Both run entirely inside your browser using WebAssembly, so nothing is uploaded.
How well it works depends very much on the style of the Urdu letters:
| Style of Urdu | Where you see it | Result here |
|---|---|---|
| Naskh style | Websites, phone screens, WhatsApp, Word with Arial or Times New Roman, many government forms | Excellent: our test images were read without errors |
| Nastaliq | Newspapers, books, InPage documents, Jameel Noori Nastaleeq | Rough draft: about 3 in 4 letters right, 1 in 3 words fully right |
These figures come from our own test: five Urdu passages, including a couplet by Ghalib, each rendered as an image at two text sizes in a Naskh-style font and in two Nastaliq fonts (Noto Nastaliq Urdu and Gulzar), then read by the tool and compared letter by letter with the original. Real photos with shadows, blur or curved pages will score lower.
We think it is better to say this clearly than to promise perfect Nastaliq OCR. For Nastaliq, the tool still saves typing on long texts, but plan to correct every line.
Worked example
A screenshot of this sentence in a Naskh-style font:
پاکستان کا دارالحکومت اسلام آباد ہے۔ یہ شہر انیس سو ساٹھ کی دہائی میں بسایا گیا تھا۔
is read back exactly, letter for letter. The same sentence set in Noto Nastaliq Urdu came back like this in our test:
کستان کا وارالکومت اسلا مآباد ہے۔ یہ شہرایس سو سا ھکی دبئی میں بسایاگیا تھا۔
پاکستان کا دارالحکومت اسلام آباد ہے۔ یہ شہر انیس سو ساٹھ کی دہائی میں بسایا گیا تھا۔
Most of the sentence is there, but letters are missing or confused (د read as و), spaces move, and two words are joined. Fixing it takes a minute; retyping the whole line would take longer, but on a full Nastaliq page the corrections add up.
Common problems and fixes
The result is mostly wrong
The image is probably Nastaliq, small, blurred or skewed. Crop to the text, make sure it is straight and large enough (letters at least 25 pixels tall), and try again. For long Nastaliq documents, typing with our phonetic Urdu keyboard may be faster.
Words are joined or split in odd places
This is the most common OCR error in Urdu, because spaces between Urdu words are small. Fix them in box 2, then run the text through the Urdu Spell Checker to catch the rest.
My PDF has real text, not a scan
Use Urdu PDF to Word instead: it takes the text out of the PDF directly, without OCR, so there are no reading errors.
Nothing happens on an old phone
OCR needs a modern browser with WebAssembly and some free memory. Try a current version of Chrome, Edge, Firefox or Safari, and close other tabs.
Who uses this tool
- Students turning photos of notes, notices and book pages into text they can edit.
- Journalists and researchers quoting Urdu from screenshots and scanned documents.
- Office staff retyping scanned applications and forms.
- Anyone who received Urdu as a WhatsApp image and needs the words as text.
Private by design
The OCR engine and Urdu model are downloaded to your browser from this site the first time you use the tool. After that, everything happens on your device: your images and the text in them are never uploaded or stored.