How OCR works and why it makes mistakes
OCR turns a picture of text into text you can copy. It works well on clean printed pages and badly on blurry photos or handwriting. This guide explains the steps inside an OCR engine so you can understand the mistakes and avoid many of them.
The steps inside an OCR engine
- Preparation: the picture is converted to black and white or grayscale so letters stand out from the background.
- Layout analysis: the engine finds blocks of text, lines and words.
- Recognition: each line is read, in modern engines by a neural network trained on many examples of letters, using the chosen language to prefer likely words.
- Output: the recognized characters are put together as text.
Which engine this site uses
Image to Text uses Tesseract, a well known open source OCR engine, running in your browser as WebAssembly. It supports English, Spanish, French and German in this tool, and each language needs its own data file, which is downloaded from this site the first time you use it.
Why mistakes happen
- Low resolution: small, blurry letters look alike, so an l, a 1 and an I get confused.
- Poor contrast, shadows or colored backgrounds make letters harder to separate.
- Tilted or curved pages distort lines and words.
- Handwriting, decorative fonts and very small type differ from what the engine learned best.
- The wrong language setting makes the engine prefer the wrong words and misses accented letters.
Prepare the picture for fewer errors
- Capture or scan at around 300 dpi for documents.
- Keep the page straight and evenly lit, without glare.
- Crop to the text so borders and images do not confuse the layout analysis.
- Choose the correct language before you add the file.
Check the result
Even a good OCR result needs a quick proofreading pass. Pay extra attention to names, dates, numbers and punctuation, where a single wrong character can change the meaning. Tables and columns come out as plain text, so layout has to be rebuilt by hand.
Frequently asked questions
Is OCR accurate?
On sharp, well lit, printed text it can be very accurate, and on poor images it can be very poor. Accuracy depends mostly on image quality and the correct language.
Can OCR read handwriting?
Engines built for printed text usually read handwriting poorly. Neat block letters sometimes work, cursive rarely does.
Is it safe to OCR private documents here?
The picture is read in your browser and is not uploaded, so the document stays on your device.
Try the converters
- Image to Text Converter (OCR)Add a photo, screenshot or scanned PDF and get the words back as text you can copy.
- PDF to PNG ConverterOpen a PDF and save each page as a PNG image.
- PNG to PDF ConverterAdd PNG images or screenshots, arrange them and create one PDF.
Last reviewed October 2026. Steps on other systems can change between versions, so check the menu names on your own device. Convert In Place is not affiliated with Apple, Google, Microsoft or Adobe.