How to Get Good Results
OCR results depend entirely on the quality of the image. A few habits make a large difference to accuracy:
- Resolution: the text should be sharp and reasonably large — very small or blurred text produces more errors. For a scan, 300 DPI is good.
- Keep it straight: a straight image reads far better than a tilted or skewed one.
- Light and contrast: black text on white paper works best; avoid shadows and glare.
- Printed text: books, newspapers and text printed by a computer read well; handwriting still cannot be read reliably.
- No language to choose: Bangla, English or a mix of both — it is all read together. There is no option to pick.
Is the image sideways? No problem — the orientation is detected automatically and straightened before reading. If you still need to, rotate it with ⟲ Left / ⟳ Right and press "Read Again".
For NID Cards, Passports and Photographed Documents
A phone photo of an NID card, a passport or a legal document is harder than an ordinary image because of the security patterns, stamps and glare. These three steps improve the result a great deal:
- Select only the part you need — drag with the mouse (or your finger on a phone) across the text in the preview below, then press "Read Again". Reading a small area instead of the whole page raises accuracy considerably.
- Bangla and English come out together — an NID or passport has the Bangla name beside the English details; both are read in one pass, with nothing to select.
- Leave image cleanup switched on — a phone photo is usually bright on one side and shadowed on the other; our processing works out the light and shade for each area separately and lifts out just the text. Turn it off only for a very clean scan.
Remember: a straight, close, well-lit photo makes any OCR several times better — a shadowed or tilted shot will always carry more errors.
The first time you use it, the engine and the Bangla model (a few megabytes) download once. After that they are cached in your browser, so later images are fast.
After You Have the Text
You can use the text straight away on this site — send it in one click to the Bangla Spell Checker (the most useful step, for catching small OCR errors), the Unicode Editor, Bangla to English, or Unicode to Bijoy for press work.
Frequently Asked Questions
Is my image uploaded to a server?
No. The whole recognition process runs inside your browser, using WebAssembly — neither the image nor the text reaches our server. So personal documents are safe here.
Can it read handwriting?
Not reliably. The engine is built for printed text — books, newspapers, screenshots and scans. Results on handwriting are very uncertain.
Can I extract text from a PDF?
Not directly — take a screenshot of the page, or save it as an image, and it will work.
Why are there some mistakes in the result?
Bangla conjuncts and vowel signs are complex, so some OCR errors are normal. Better image quality reduces them; after that, the Spell Checker catches most of what is left.