Image to Text (OCR)

Step by step: from image to text

1. Pick the language of the text. Use the Language list above. It starts on English. The tool reads one language per run, so pick the one most of your text is in.
2. Add one image. Drop it on the box or click the box to choose a file. A preview and the file size appear under the box. If you drop several files, only the first one is used.
3. Click Extract Text. A progress bar shows each stage: loading the language data, starting the engine, and reading the text.
4. Read the result. The text appears in the Extracted Text box as plain text. There is no bold, no font size and no table layout.
5. Fix and save it. You can fix mistakes right in the box. Then click Copy to Clipboard, or Download as TXT to save a file called extracted-text.txt. Both use the text as it is in the box, with your changes.

What runs on your device, and what is downloaded

The tool uses Tesseract.js, a free, open-source OCR engine (OCR means optical character recognition, or reading text from a picture). Your image is read by your own browser. It is not sent to FreeToolPoint or to anyone else.

The engine itself is not stored on this site. The first time you click Extract Text, your browser downloads the engine program and the data file for the language you picked from the jsDelivr CDN, a public file host. Each language file is about 1 to 3 MB. On a slow connection this first run can take a while. Your browser usually keeps a copy, so later runs in the same language start much faster. If you pick a new language, that language file is downloaded the first time you use it.

What OCR reads well, and what it struggles with

OCR works best when the image looks like a clean printed page. The further your image is from that, the more mistakes you should expect.

Tables are a special case. The words in each cell are often read correctly, but the grid is lost. A row comes out as one line of words with spaces between them, and you have to rebuild the columns yourself.

Prepare your image before you upload

A few minutes of preparation often saves more time than fixing the text afterwards.

• Crop to the text. Remove borders, desk edges, photos and logos. Less noise means fewer strange characters in the result. You can use the Crop Image tool. For a page with two columns, crop and read each column on its own.
• Hold the camera straight. Take the photo from directly above the page, not at an angle. Lines of text should be level, not sloping.
• Use even light. Daylight from a window works well. Avoid your own shadow and the phone flash on glossy paper.
• Make sure the text is big enough. A rough check: open the image and zoom to 100%. If you can read the smallest words easily, the engine has a good chance. If you have to squint, it will struggle too. For a very small screenshot, making it two times larger with Resize Image can help, but it cannot add detail that was never there.
• Keep good contrast. Dark text on white or light paper is best. Grey text on a coloured background gives weaker results.
• Use a common file type. JPG, PNG, WebP and BMP work best. Other types your browser can open, such as GIF, are changed to PNG first. Some phones save photos as HEIC. Most browsers, except Safari, cannot open HEIC, and the tool shows a message when that happens. Save those photos as JPG or PNG first.

Check and fix the result

OCR is rarely perfect. Always read the result next to the original image, most of all for numbers, names, dates and prices. The tool does not show a score for how sure it is about each word, so you need to check yourself.

The same mix-ups come up again and again, because some characters look almost the same:

• 0 and O (zero and the letter O)
• 1, l and I (one, small L and capital i)
• rn and m, so turn can become tum
• 5 and S, and 8 and B
• Commas and full stops, which matter a lot in prices

For example, a receipt line that says Total 110.00 might come out as Tota1 1l0.O0. The words look close enough to miss, but the number is wrong. A good habit is to search the text for the letters O and l inside numbers.

You can fix mistakes right in the Extracted Text box. Click in the box and type. Copy to Clipboard and Download as TXT use the text with your changes. If you click Extract Text again, the box gets a new result and your changes are lost, so copy or save your work first.

If the tool says no text was found, check that the right language is selected. Then try a sharper, tighter crop of the text.

About the language list

The list has 10 languages: English, Spanish, French, German, Portuguese, Chinese (Simplified), Japanese, Korean, Hindi and Arabic. Other languages are not offered here. If a page mixes two languages, pick the main one. Words in the other language may come out wrong. Picking English for a Chinese or Arabic image will give nonsense, because the English model does not know those characters.

Getting text out of a scanned PDF

This tool only accepts images. A scanned PDF is really a set of page images, so it needs two steps. First, turn each page into an image with PDF to Image. PNG at 2x or 3x scale keeps small text sharp. Then upload the pages here one at a time. If you can already select and copy text in your PDF viewer, the PDF has real text inside. In that case you do not need OCR. Just copy the text directly.