Extracting text from images and scanned PDFs: how OCR works and how to improve it
Optical character recognition (OCR) reads the shapes of letters in a picture and turns them into text you can copy and edit. This guide explains what to expect, how to get clean results, and how to work with Hindi and other languages.
Common uses
Copy text from a photo of a page
Photograph a notice, a book page or a bill and paste the words into a message or document.
Scanned documents
Pull text out of scanned certificates, letters and forms so you can search or quote them.
Screenshots
Copy text from a screenshot of a chat, a video slide or an app that does not allow selecting text.
Hindi and regional documents
Recognise Devanagari, Tamil, Telugu, Bengali and other scripts from printed material.
Study and research
Turn textbook photos and lecture slides into text for notes and translation.
Business paperwork
Digitise invoices and receipts to search for a number or a vendor name later.
What OCR can and cannot do
Modern OCR is very accurate on clean, printed text and weaker on handwriting, stylised fonts, low resolution and curved or shadowed pages. Treat the output as a fast first draft and proofread important details.
Choosing the language
Each language has its own alphabet and spelling patterns, so recognition improves a lot when you choose the right one. English is always read together with your choice, which suits mixed documents such as Hindi forms with English names and numbers.
Photo tips
Hold the phone parallel to the page, use even light, fill the frame and keep text horizontal. Tap to focus and avoid flash glare. A flat scan at 300 dpi is better than a tilted photo.
PDFs
If a PDF already contains real text, it is copied directly and instantly. Pages that are pictures are recognised page by page, up to 15 scanned pages at a time.
Output
You get plain text in reading order that you can copy or download as a .txt file named with the time. Tables become lines of text rather than cells.
Privacy
The picture is deleted as soon as its text is read, and the text 2 minutes after you are done; nothing is kept or shared.
Troubleshooting
The text is garbled
Most often the wrong language is selected. Choose the language of the page and try again.
Some words are missing
Retake the photo closer, straighter and in better light, or increase the scan resolution.
Numbers look wrong
Proofread digits such as 0/O and 1/l, especially in amounts, dates and ID numbers.
Available in 55 languages
The whole site - menus, buttons, guides and messages - can be switched to 55 languages, including English, Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Odia, Assamese, Nepali, Urdu, Arabic, Spanish, French, German, Portuguese, Italian, Russian, Turkish, Japanese, Korean and Chinese. Use the language menu in the top bar; your choice is remembered, and the file formats and tools work exactly the same in every language.