Extract Text From an Image
How to Get Text From an Image
A photo of a document, a scanned page or a screenshot, up to 100 MB.
Pick the language the text is written in. This decides whether letters such as š, ž, ü or ç are read correctly.
You get a plain text file, or a PDF that looks like the picture but has text you can select and search.
What OCR does, and how well
OCR (optical character recognition) looks at the picture, finds the lines and letters and writes them out as text. It is done on our server by Tesseract, the open-source recognition engine. Clean, printed text is read almost without errors; the further a picture is from that, the more corrections you will need.
What to expect from different pictures
| Picture | Result |
|---|---|
| Screenshot, scan or sharp photo of a printed page | Very good. Usually only a few errors per page |
| Photo taken at an angle, with shadows or a curved page | Usable, with errors where the text is distorted or dark |
| Tables and text in several columns | The words are right, but the order of lines and cells may be mixed |
| Receipts on thermal paper, small or faded print | Partly readable; check every number |
| Handwriting | Not supported. The engine is built for printed letters |
The language setting matters more than anything else
The engine does not guess the language; it uses the one you choose to decide between similar-looking letters. In our test with a Bosnian sentence, the setting "Bosnian, Croatian" returned every word correctly: Račun, Šećer, čaj, žito, đak, ćevapi. With "Mixed or not sure" the same picture came back as Racun, Zito and dak. Use the mixed setting only when a page really contains several languages.
The page pre-selects a language from your browser settings. Change it if the document is in a different one. English is always recognised alongside the language you pick.
Getting a better result
- Photograph straight from above in even light, with the page filling the picture. Crop away the table around it with the crop tool.
- Turn the picture the right way up first; text lying on its side is not read. Rotate Image does that.
- Resolution: letters should be at least about 20 pixels tall. Very small screenshots are enlarged automatically before reading; pictures larger than 6000 pixels are reduced.
- Dark or low-contrast photos improve with more contrast in the Image Editor.
Text file or searchable PDF?
- Text file: only the words, line by line, in UTF-8. For pasting into a document, an e-mail or a translation tool.
- Searchable PDF: the picture stays as it is, with the recognised text placed invisibly behind it. You can search, mark and copy text while still seeing the original page.
Good to know
- One image per run. For a scan with many pages, use OCR PDF.
- If no letters are found at all, you get a message instead of an empty file.
- The picture is uploaded over HTTPS, processed on our server and deleted within 4 hours. For an editable Word document from a scanned PDF, see Scanned PDF to Word with OCR.
Frequently Asked Questions
Is image to text free?
Yes. No registration, no watermark and no limit on the number of images.
Which languages are supported?
English, Bosnian, Croatian, Serbian (Latin and Cyrillic), German, French, Italian, Spanish, Russian and Turkish.
Can it read handwriting?
No. The recognition engine is made for printed text. Handwriting gives mostly wrong results.
Why are letters like š, ž or ü wrong?
The language setting does not match the text. Choose the language the document is written in and run it again.
Does it keep the layout?
The text file keeps the lines but not fonts, columns or tables. The searchable PDF keeps the full look of the page.
How long is my image kept?
The upload and the result are deleted from the server within 4 hours.