In this article
Someone sends you a screenshot of an address. You photograph a printed form to fill in later. You have a scanned page of notes that will not let you copy a single sentence. In each case the words are there, but to your computer they are just a pattern of dots, not text.
The Image to Text (OCR) tool reads the text in a photo, screenshot or scanned page and lets you edit it, copy it, or save it as a text file. This guide explains how to prepare the picture, how to pick the language, and what mistakes to look for.
What OCR actually does
A picture of text is made of pixels, not letters. OCR software finds lines of text, splits them into shapes, and decides which letter each shape most likely is, using what it knows about the language. The tool uses the open-source Tesseract engine, which runs inside your browser. It never “understands” the meaning; it matches shapes, so the cleaner the shapes, the better the result.
- 1Add the pictureA photo, screenshot or scanned page. The tool takes one picture at a time.
- 2Choose the languageThe language in the picture, not the language you want back.
- 3Extract textThe first time, the language data is loaded from this site, which can take a moment. After that it is quicker.
- 4Check and editThe result appears in a box you can type in, so you can fix mistakes right there.
- 5Copy or downloadCopy the text, or save it as a
.txtfile named after your picture.
Your picture is read on your own device. Only the language data is downloaded, and only once.
Choosing the language
OCR needs to know what to expect. The tool offers English, Hindi, English + Hindi, Spanish, French, German, Arabic, Bengali, Tamil, Telugu, Marathi, Gujarati, Punjabi and Russian.
| Your picture contains | Choose | Why |
|---|---|---|
| Only English text | English | Fastest and most accurate for plain English |
| Only Hindi in Devanagari | Hindi | Uses Hindi letter shapes and the headline bar |
| Hindi and English together, such as forms and bills | English + Hindi | Reads both scripts in the same picture |
| Marathi (also in Devanagari) | Marathi | Designed for Marathi words and spellings |
| Tamil, Telugu, Bengali, Gujarati or Punjabi | The matching language | Each script has its own letter shapes |
Picking the wrong language is the most common cause of garbled output. If the result is nonsense, try another choice.
Getting the best result: prepare the picture
A few minutes of care with the picture beats any amount of correcting afterwards.
- Make the text big enough. Very small text, such as a screenshot shrunk to fit, loses detail. Crop in on the part you need and use a larger original where you can.
- Keep it sharp. Hold the phone steady, tap to focus and avoid zooming digitally.
- Shoot straight on. Tilted pages and curved book spines bend the lines and confuse the reader.
- Use even light. Shadows across the page and glare from glossy paper hide letters.
- Contrast helps. Dark text on a plain light background is ideal. Text over photographs, patterns or coloured gradients is much harder.
- Crop out the clutter. Borders, pictures and stamps around the text can add stray characters. The Image Cropper trims the picture first.
- Likely to fail0–2
- Needs proofreading2–4
- Usually reads well4–5
Give one point each for: big text, sharp focus, straight angle, even light and plain background. This is a rule of thumb, not a measured score.
What OCR still gets wrong
- Handwriting. Neat printed block capitals sometimes work. Ordinary cursive handwriting mostly does not.
- Look-alike characters. The letter O and the digit 0, the letter l and the digit 1, and rn read as m. Check numbers such as phone numbers, amounts and ID numbers carefully.
- Hindi marks and joined letters. Small vowel marks and conjunct letters can be dropped or changed. Read Hindi output through once against the original.
- Tables and columns. The words come out, but the layout does not. Columns can run together and table cells can end up in a different order.
- Fancy fonts and decorative text. Logos, stylised headings and curly typefaces are often missed.
If the tool says it found no text, the picture is usually too small, too blurry or in a different language from the one selected. Try a clearer, larger picture and the right language.
A scanned PDF instead of a picture
If your text is inside a scanned PDF, turn each page into a picture first with the PDF to Image Converter, then run the pictures through the text reader one by one. A scanned PDF is just a stack of photographs, which is why you cannot select the words in it. The same reason makes scanned PDFs so large, as explained in how to compress a PDF.
A privacy note
Pictures you want to read are often personal: bills, forms, IDs, letters. The tool reads your picture on your own device. The only thing downloaded is the language data, which comes from this site and is stored once. If you share the result, remember that a text file may contain personal details such as addresses and numbers.
Text read from an image often arrives in odd capitals. Fix it with the Case Converter, and see how to change the case of any text.
See also: turning typed text into handwritten pages.
See also: joining the short lines OCR text often comes with.
Questions people ask
How do I extract text from an image?
Open the Image to Text (OCR) tool, add the picture, choose the language in the image and press Extract text. The words appear in a box where you can edit them, copy them or download a text file.
Can it read Hindi text from a photo?
Yes. Choose Hindi for pure Hindi, or English + Hindi for bills, forms and notices that mix both. Check the result against the original, because small vowel marks and joined letters can be misread.
Why is the text I got garbled?
The language was probably wrong, or the picture was blurry, tilted, very small or low in contrast. Choose the right language, crop in tightly and try a sharper, straighter picture.
Can OCR read handwriting?
Not reliably. It works best on printed text. Neat block capitals sometimes read well, but ordinary joined-up handwriting usually does not.
How do I copy text from a scanned PDF?
A scanned PDF holds pictures of pages. Convert the pages to images with the PDF to Image Converter, then run each image through the OCR tool and copy the text.
Is the text reader accurate for numbers?
Mostly, but it can confuse look-alike characters such as O and 0 or l and 1. Always compare phone numbers, amounts and ID numbers with the original.
Which languages are supported?
English, Hindi, English + Hindi, Spanish, French, German, Arabic, Bengali, Tamil, Telugu, Marathi, Gujarati, Punjabi and Russian.
Is it safe to use an online OCR tool?
It is safest when the picture stays on your device. The Image to Text tool on Toolcloak reads your picture in your browser. Only the language data is downloaded, once.