Image to Text OCR
Extract English text from images locally with OCR.
Extract English text from images locally with OCR.
Optical character recognition, usually shortened to OCR, reads the shapes of letters in a picture and turns them into editable text. This tool applies it to a single image: choose a screenshot, a scan of a printed page or a photo of a sign, press Extract text, and the recognized words appear in a box below the button.
The recognizer is set to English. That means it expects Latin letters and English words, and it will perform poorly on pages in other languages or scripts. Accuracy depends heavily on the picture. Sharp, straight, high-contrast printed text gives good results, while handwriting, curved pages, glare, heavy shadows or tiny fonts produce errors that you will need to correct.
While recognition runs, the button reads Reading and is disabled, and large images can take noticeable time on slower devices. When it finishes, the text is shown as plain text, so you can select it and copy it elsewhere. Treat the output as a first draft: skim it against the picture, especially numbers, names and punctuation.
Error messages, chat snippets and slides often come as images you cannot select. Run the screenshot through OCR and paste the recognized words into a note, search box or bug report instead of retyping them.
A photographed or scanned page of a letter, recipe or handout can become editable text. Flat, well-lit pictures taken straight on work best, and you can fix the few mistakes yourself.
Phone numbers, addresses or opening hours on a poster or label are easier to save as text. Crop tightly first so the recognizer ignores background clutter and focuses on the words.
Text locked inside images cannot be searched in your notes app. Extract it and store the words alongside the picture, so a future search by keyword actually finds what you saved.
Recognition runs in your browser, and the image is not intentionally sent to an ELDOSA server. The OCR engine may need to fetch its own support files the first time it runs, so a connection can be required.
It is currently set up for English only. Pages in Spanish, German or other languages written in Latin letters may be partly readable, but accents and special characters will often be wrong. Scripts such as Arabic, Chinese or Cyrillic are not supported by this configuration.
OCR works by guessing letters from pixel patterns, so blur, low contrast, odd fonts, shadows or skewed angles all cause errors. Try a sharper, straighter and tighter-cropped image. Even with a good picture, proofreading is wise, particularly for numbers and proper names.
Not dependably. The recognizer is designed for printed text, and handwriting varies too much from person to person. Neat block capitals written in dark ink on plain paper may give partial results, but expect to retype most of it.
No. The output is plain text in the order the engine detected it. Columns may be merged or interleaved and tables lose their grid. If layout matters, crop each column or section separately and run them one at a time.
A flat, evenly lit, high-contrast picture of printed text works best: dark letters on a plain light background, photographed straight on with the text filling most of the frame. Screenshots of on-screen text are usually cleaner than photos of paper.