Image to Text

What happens to your image

Image to text, also called OCR (optical character recognition), looks at the shapes of the letters in a picture and works out which characters they are. Here that work is done by an open-source OCR engine, Tesseract, running inside your browser tab. The first time you use a language, your browser downloads that language's model from this site, between one and three megabytes. After that the image is read on your own device.

Nothing about the picture is sent anywhere: not to DocTranslator, not to an OCR service. That matters for the things people most often photograph to read, such as a letter from an office, a prescription, a contract page or a passport.

How to get clean text from a photo

OCR is only as good as the picture it reads. Four things make the biggest difference:

• Choose the right language first. The engine reads with the alphabet and vocabulary of the language you pick. Russian read as English comes out as nonsense, and accents in French or Polish go missing when the language is wrong.
• Photograph the page flat and square. Hold the phone directly above the paper, fill the frame with the page, and avoid a curl near the spine of a book. Lines that bend confuse the engine more than a small picture does.
• Light it evenly. A shadow across the page, or the glare of a flash on glossy paper, hides letters. Daylight from a window, without the flash, usually works best.
• Keep it sharp. Tap the screen to focus on the text before you take the photo. Small print needs to be large enough in the picture to read with your own eyes.

What OCR gets wrong

Even a good scan comes back with a few errors, and they tend to be the same ones: the letter l read as the number 1, O as 0, rn as m, a comma lost at the end of a line. Names, numbers and dates deserve a second look before you copy them into anything that counts.

Handwriting is much harder than print, and joined-up handwriting is often read badly or not at all. The output is plain text, so the layout of the page is not kept: columns are read one after the other, and a table comes back as lines of words. When you need the layout too, the next section is the way to keep it.

When you need the text in another language

Copying the text into a translator loses the page: the stamps, the table, where each line sat. To translate a scanned document with its layout, upload the scan to DocTranslator instead. It reads the scan with OCR, translates the text and returns a file that looks like the original, in any of 108 languages.

The browser tool on this page reads English, Spanish, French, German, Italian, Portuguese, Russian, Ukrainian, Polish, Turkish, Arabic, Hindi, Chinese (Simplified), Japanese, Korean and Vietnamese. In a scanned PDF it reads the first 5 pages; a translation reads every page, in every language DocTranslator supports.