OCR PDF

OCR PDF reads the words in scanned pages — from a scanner, a phone camera or a fax — and makes them real text. Choose a searchable PDF, which keeps every scanned page exactly as it was and lays an invisible layer of the recognised words over it, so the file can be searched, and its words selected and copied; or a plain text file. It reads English, ten Indian languages (Hindi, Marathi, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi and Urdu) and nine more, up to three in one document. Recognition runs in your browser with Tesseract, the open-source engine, and the result tells you how confident it was.

1. Choose a scanned PDF, or drop scans and photos on the page: images become pages too.
2. In OCR, tick the languages the document is written in. English is ticked; your browser's language may be too.
3. Choose a searchable PDF or a text file. Pages that already have text are skipped unless you untick that.
4. Press Read. The first time, the reading engine and the languages are downloaded — the panel says how much — and kept by your browser for next time.
5. Each page takes a few seconds. If any line was read with doubt, those lines are shown next, each under the strip of the scan it came from, with the unsure words marked: correct what is wrong, then make the file.
6. Download it. The result says how confident the reading was and which pages you corrected.

• OCR reads clear printed text well. Handwriting, very small or decorative print, pages photographed at an angle, blur and low resolution all bring mistakes. A confidence under 70% means check the text carefully.
• Tick only the languages actually in the document: an extra one slows reading and can turn words into look-alikes from the other script.
• Tables and columns come out as lines of text in reading order as the engine sees it, not as a table.
• Speed depends on the device: a few seconds a page on a laptop, longer on a phone. A 100-page scan takes a while; you can cancel at any point.
• The first run downloads 3 to 8 MB (the engine and one to three languages). Files up to 300 MB each; a password-protected PDF asks for its password.
• Your files are read in this browser tab and are never uploaded to ToolsDev or anyone else; nothing is saved when you close the page. To read, the page downloads the Tesseract engine and the language models you choose from toolsdev.in — none of those requests carries your document.

No. In the searchable PDF the pages are exactly the original pictures; the recognised words are an invisible layer on top, placed over the words they came from.

Yes: Hindi, Marathi, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi and Urdu, each with Tesseract's trained model. Accuracy is best on clean print; matras and conjuncts in small or faded type are where errors appear.

No. Only the reading engine and language files come from the server, the same for everyone. Your pages are read on your own device.

The engine guesses each word from the picture. Smudges, shadows, skewed pages and unusual fonts make it guess wrong. Rescanning at 300 dpi, flat and straight, helps most.

Yes, before the file is made. Lines with a word read at under 70% confidence are listed with the part of the scan they came from; tick Show every line to see the rest. A corrected word stays over its word in the scan when the line keeps the same number of words; otherwise the line's words share its width, so they still search and copy correctly.