OCR image to text

OCR image to text reads the printed words in a photo, a scan or a PDF and gives them back as text, in English, Hindi or both. Copy the text or save it as a file. It is for the moments when you need the words from a notice, a letter or a scanned page and retyping them would take too long. Everything runs in your browser with Tesseract, an open-source OCR engine.

How to use

1. Drop one photo, scan or PDF on the box, or click it to choose it. PNG, JPG and WebP pictures work, and so do PDFs.
2. Under Language of the text, choose English, Hindi, or English and Hindi for a page that mixes both. Only the language you choose is downloaded.
3. Press Read text. The first time, the OCR engine and the language data download first; the bar shows each step, then each page as it is read.
4. Check the text in the box, then press Copy text, or Download .txt to save it as a file. Cancel stops a long PDF.

Examples

A clean test picture of two printed lines, black on white, read with English chosen, came back exactly as printed: “The quick brown fox jumps” on the first line and “over the lazy dog.” on the second. A picture of two lines of printed Hindi, read with Hindi chosen, came back as “भारत एक बड़ा देश है।” and “हिंदी भाषा सुंदर है।”. A two-page PDF gives the text of each page under a heading, Page 1 and Page 2. Notice that these were sharp, flat pictures made on a computer; a photo of paper taken with a phone can come back with wrong letters, so check it.

Limits

The result depends on the photo. Sharp, straight, well-lit printed text reads best; blur, shadows, slanted pages, small print and decorated fonts lead to wrong letters, so always check the text before you use it. Hindi is read less well than English. The tool is made for printed text, not handwriting. Tables, boxed cells and pages that mix columns, pictures and text often come out wrong: words can land in the wrong order or run across cells. Even on a clean page, small letter errors can happen, such as a 0 read as an O or a 1 read as an l.

One file at a time, up to 20 MB, as Quick facts shows, and a PDF of up to 20 pages; split a longer PDF first. A photo larger than 4,000 pixels on its longer side is scaled down to 4,000 before it is read, and PDF pages are drawn at 300 dpi within the same size. The first use downloads the OCR engine, up to 3.5 MB, and the language data you choose: English 2.8 MB, Hindi 1.3 MB. After that your browser can reuse its copy, unless its cache is cleared. Columns, tables and page layout are not kept: the text comes back as plain lines. Phones have less memory and are slower, so a long PDF can take minutes there, or fail before these limits.

FAQ

Is my file uploaded to read the text?

No. Your browser reads the file and runs the OCR engine on this page. The engine and the language data come from this site, and your photo or PDF never leaves your device.

Why is some of the text wrong?

OCR guesses each letter from its shape, so anything that blurs or bends the letters causes mistakes. Retake the photo straight on, in good light, with the page filling the frame, and try again.

Can it read a page that mixes English and Hindi?

Yes. Choose English and Hindi, and both language files are used together. This downloads both files the first time, and reading takes longer than with one language.

Does it keep the layout of the page?

No. The result is plain text, line by line. Tables, columns, bold text and pictures are not kept, so a page with columns may come back with the lines of the columns mixed.