How to OCR a scanned Arabic PDF and make it searchable
A scanned document is a picture of a page. You can read it, but your computer cannot: searching for a word finds nothing, copying gives nothing and an editor has no text to change. Optical character recognition (OCR) reads the picture and adds the text it finds to the file.
Arabic is one of the harder scripts for OCR, because letters join, change shape and carry dots and small marks that are easy to lose on a poor scan. This guide shows how to get a good result and how to check it. Everything happens in your browser: the scan is never uploaded.
Before you start: get a clean scan
The quality of the picture matters more than anything else. A few minutes spent on the scan saves a lot of corrections later.
- Scan at 300 dpi when you can. Lower resolutions lose the dots that tell many Arabic letters apart.
- Keep the page straight and flat. A photo taken at an angle, or a curved book page, gives much weaker results.
- Prefer black text on a white page with even light, without shadows or a coloured background.
- If you only have photos, turn them into a PDF first with JPG to PDF: one page per picture, turned the right way.
Step by step
- Open the scanned PDF in the OCR PDF tool. PDFancy notices that the pages have no text and offers to recognise it.
- Click Recognise text. In Document languages, tick Arabic. If the page also contains French or English words, tick that language as well; otherwise untick it, which makes recognition faster and more accurate. Arabic is ticked for you when your browser is set to Arabic.
- Start the recognition. The dialog shows which page is being read. It takes a few seconds per page, on your device.
- When it is done, every page carries its recognised text. Click a paragraph to check it or correct it.
- Download the PDF. It looks exactly like the scan, but the text can now be searched, selected and copied.
What you get
The recognised text is added as an invisible layer exactly over the words of the picture, in reading order. That is what makes the document searchable in any PDF reader and readable by screen readers, while the page itself keeps the look of the original scan.
You can also correct or change the text. When you edit a recognised paragraph, the old words are covered with the colour of the page background and the new words are drawn on top, right to left, in the bundled Arabic font closest in style: Amiri for pages that look classic and serif, IBM Plex Sans Arabic for plainer pages. This works best on clean white pages. For a deeper look at editing, read how to edit an Arabic PDF.
Finally, the recognised document can go to Word: see how to convert an Arabic PDF to Word.
Limits to know
- The engine reads printed text. Handwriting and ornate calligraphy, such as Thuluth or Diwani titles, are rarely recognised reliably.
- Vowel marks (harakat) are often missed or simplified, so fully vocalised texts such as poetry need careful checking.
- Names, numbers and words in rare fonts are the most likely to contain mistakes.
- For Arabic paragraphs the app does not try to identify the exact typeface of the scan; it picks one of its two Arabic fonts by style.
How to check the result
- Use the reader’s search (Ctrl+F) to look for a few words you can see on the page, including a name and a number.
- Select a paragraph, copy it and paste it into a text editor to read it as plain text, from right to left.
- Compare important figures, such as amounts and dates, with the picture before you rely on the text.
Tips
- Tick only the languages that really appear in the document: each extra language slows recognition and adds chances of confusion.
- Recognise a single test page first when a document is long, to check that the scan quality is good enough.
- Keep the original scan: the recognised PDF is a new file, and you may want to try again with a better scan later.
- Recognition runs on your device, so a recent computer is faster than an old phone for long documents.
Frequently asked questions
Is my scanned document sent to a server for OCR?
No. The recognition engine and the Arabic language data are part of the app and run in your browser. Nothing is uploaded.
Which languages can be recognised?
Arabic, French and English, alone or together on the same page.
Can it read handwritten Arabic?
Not reliably. The engine is made for printed text; handwriting usually gives poor results.
Will the PDF look different after OCR?
No. The page keeps the picture of the scan; the recognised text is added invisibly on top, until you choose to edit a paragraph.