A scanned contract from which you want to quote a paragraph, a photographed invoice whose number must be copied, an old PDF book where Ctrl+F finds nothing: in all these cases the “text” is only a picture. Optical character recognition (OCR) turns it into real text. It is never perfect, but well prepared it saves hours of typing.
How OCR works, in two minutes
OCR software splits the image into lines, then words and letters, and compares each shape with what it has learned. It gives each word a confidence score. When the image is sharp, contrasted and straight, most words are recognised with certainty; when it is blurry, tilted or noisy, errors multiply and the score drops.
In MaxyTools, recognition takes place in your browser: language models are downloaded once, then the image you process never leaves your device. That matters for documents such as an ID, a prescription or a contract.
Which tool for which case
| Your file | Tool | What you get |
|---|---|---|
| A photo, screenshot or scan (JPG, PNG, WebP, BMP, GIF, HEIC) | Image OCR | Editable text, doubtful words highlighted, export to .txt, searchable PDF (Latin text) or ZIP. |
| A PDF whose pages are images | OCR PDF | A .txt and a searchable PDF where Ctrl+F and text selection work; up to 150 pages per run. |
Nine languages are available (French, English, Arabic, German, Italian, Dutch, Norwegian, Danish and Swedish), and you can tick several at once (for example French and English).
Photographing or scanning for good results
Nine times out of ten, recognition quality is decided before the tool. A few reference points:
- Lay the page flat, facing you, with no shadow from your phone. Even lighting beats a flash that creates glare.
- Aim for good resolution. A lowercase letter should be at least twenty pixels tall. At 300 dpi, 10-point printed text meets that; at 150 dpi it is borderline for small type.
- Contrasting background, dark text. Light grey text on a cream background is a problem; the tool's adaptive black and white helps.
- Avoid creases, stamps over text and photos taken at an angle.
The tool offers optional settings: straightening, ×2 enlargement for small characters, greyscale, contrast and adaptive black and white. Try them one at a time rather than all at once.
Making a scanned PDF searchable, step by step
- Open OCR PDF and drop the PDF (100 MB maximum, unlocked).
- Choose the language (or several), the pages to process and the reading resolution (150, 200 or 300 dpi). 200 dpi is a good compromise; 300 dpi for small characters.
- Start: progress is real, and you can cancel and resume where you left off. Count a few seconds per page.
- Download the text (.txt) and the searchable PDF: your original pages are neither altered nor re-encoded, the recognised text is inserted on top, invisible.
- Proofread the words highlighted as doubtful, especially digits, proper names and amounts.
If the PDF already contains text, the tool tells you and can skip those pages.
Common mistakes
- Trusting OCR for digits without proofreading: a 5 read as a 6 in an amount has consequences.
- Forgetting to tick the right language: French accents read with an English model give strange words.
- Processing a PDF that already contains text: pointless, and it can make the file heavier.
- Expecting OCR to read handwriting: that is not its job and results are poor.
Honest limits
- OCR is never perfect: a sharp page gives excellent results; a blurry photo, glare, handwriting or a stamp over the text degrade them a lot.
- Arabic is harder than Latin scripts: letter dots, look-alike letters and punctuation are the most common errors, and short vowel marks (tashkeel) are not recognised. Calligraphic fonts do not work.
- The searchable PDF holds only Latin text and digits (standard PDF fonts have no Arabic glyphs): for Arabic, use the .txt file.
- Processing is slow (a few seconds per page) because it happens on your machine, not on a server.
Frequently asked questions
Is my document sent to a server to be read?
No. Recognition happens in your browser; only the language model is downloaded, never your image.
What accuracy can I expect?
On sharp, well-lit printed text, very good; on a blurry photo or handwriting, poor. Each word has a confidence score: proofread the highlighted ones.
Can several languages be read in one document?
Yes, tick them all (for example French and English) before starting recognition.
Why does the searchable PDF not contain Arabic?
Because standard PDF fonts have no Arabic characters. The .txt file does hold all the recognised text.
How do I copy the text of a scanned PDF?
Run it through OCR PDF, then open the searchable PDF: you can select and copy the text, or take the .txt file.