You have a photo of a book page, a scanned official letter or a screenshot of a post, and you need the actual text — to copy, edit, translate or quote. Retyping it could take an hour. Optical character recognition (OCR) reads the image and gives you the text in a minute.
But anyone who has run OCR on Arabic knows the results can be excellent one day and terrible the next: missing letters, misplaced dots, words fused together. This guide explains why, how to prepare images for the highest accuracy, and when OCR is the wrong tool altogether. Most of the advice applies to English and other languages too.
- Use OCR — extract text and choose Arabic only for pure Arabic, Arabic + English for mixed text.
- Best accuracy comes from straight, sharp, evenly lit images at roughly 300 DPI.
- Clean printed fonts read very well; handwriting and calligraphy read poorly.
- If the PDF already contains real text (you can select words), skip OCR and use PDF to text.
- Always proofread, especially numbers and names.
What is OCR and how does it work?
OCR stands for Optical Character Recognition. The software sees an image as coloured dots and doesn't know there's text in it. So it works through stages:
- Clean-up: convert to black and white, remove noise, correct skew.
- Layout analysis: find text regions, images and tables; order columns and lines.
- Segmentation: split lines, then words.
- Recognition: match shapes against patterns learned from huge amounts of text. Modern engines like Tesseract, used by many tools, rely on neural networks that read whole lines rather than individual letters.
- Language correction: use knowledge of the language to choose the most likely word when unsure.
Each stage depends on the previous one. If the image is skewed or blurry, line splitting stumbles and errors pile up to the end.
Why Arabic OCR is harder than English
Letters connect. Printed English letters are mostly separate, so splitting a word into letters is easy. Arabic is cursive by nature, and each letter's shape changes with its position in the word.
Dots carry meaning. At the start of a word, ب ت ث ن ي differ only by the number and position of dots. One small dot lost to compression or low resolution changes the whole word.
Diacritics. Short-vowel marks (fatha, damma, kasra, shadda) are tiny marks above and below letters that engines can confuse with dots — or ignore.
Font variety. From simple Naskh to Ruq'ah, Diwani and Thuluth, some styles stack and overlap letters in ways that are very hard for machines.
Direction and mixed text. Arabic runs right to left, while numbers and English words inside it run left to right, and the engine has to order all of it correctly.
So the same image yields lower accuracy for Arabic than English, and good image preparation matters even more.
Do you even need OCR? The 10-second test
Not every PDF needs OCR. Open the file and try to select a word:
- If you can select word by word, the PDF contains real text. Use PDF to text or PDF to Word — it's more accurate than OCR because it extracts original characters instead of guessing.
- If the whole page highlights as one block, or nothing highlights, it's an image and you need OCR.
- If you can select text but it pastes as gibberish or disconnected letters, the PDF has text with broken encoding, and OCR is the reliable fix. We cover this in Convert Arabic PDF to Word without broken letters.
How to extract text with OCR
- Open OCR — extract text.
- Choose an image (JPG or PNG) or a scanned PDF.
- Pick the text language: - Arabic only: for pure Arabic — highest accuracy, because the engine won't try to read shapes as Latin letters. - Arabic + English: for documents mixing both, like technical papers or invoices. - English only: for English text.
- Press Extract text. The first run downloads language data, so it may take a moment; later runs are faster.
- Copy or save the text, then proofread it.
Processing happens on your device — the image itself is never sent to a server. The only download is the language "training" data.
Prepare the image for maximum accuracy
This section is the difference between accurate output and a rewrite.
Resolution: 300 DPI for small print
On a scanner, choose 300 DPI, the commonly recommended resolution for OCR. Lower loses dots and diacritics; 600 DPI helps with very small print but slows processing with little gain for normal text.
With a phone, move closer so the page fills the frame — don't shoot from a distance and crop.
Straightness
Even a few degrees of skew confuse line detection. Shoot from directly above, and use your phone's document scanner to correct perspective. If a page is sideways or upside down, rotate it first with Rotate PDF. See How to scan documents with your phone.
Contrast and lighting
- Black text on white is ideal.
- Shadows over part of the page make the engine treat that area as "black" and lose the text.
- Coloured backgrounds, ornaments and watermarks reduce accuracy. Converting with PDF to grayscale can help.
Don't compress before OCR
Heavy compression creates artefacts around letters and can merge or erase dots. Use the original or least-compressed image, and compress the file after extracting text if needed.
Crop
Remove irrelevant areas — table edges, your fingers, the facing page of a book. The engine may try to "read" them and add random characters.
What reads well and what doesn't
| Text type | Expected accuracy | Notes |
|---|---|---|
| Modern print in simple fonts (Arial, Tahoma, Simplified Arabic) | Very high | The ideal case |
| Printed books in traditional Naskh | High | Heavy diacritics can affect it |
| Newspapers with multiple columns | Medium to high | Column order can get mixed |
| Fully vocalised text (Qur'an, poetry) | Medium | Diacritics may be lost or confused with dots |
| Low-resolution screenshots | Medium | Zoom in or capture at higher resolution |
| Neat Ruq'ah handwriting | Low to medium | Varies hugely between writers |
| Diwani, Thuluth or decorative calligraphy | Very low | Overlapping letters don't suit OCR |
| Faded or stained old documents | Low | Boost contrast first |
Proofreading: where errors hide
Even at 98% accuracy, a 300-word page means about six errors. Focus on:
- Numbers: 3 vs 8, 1 vs 7, and Arabic-Indic digits (٣ ٢ ٧ ٨). A wrong phone number, amount or date is worse than a typo.
- Proper names: the engine has no dictionary of names and guesses letter by letter.
- Letters that differ only by dots: ب/ت/ث/ن/ي, ج/ح/خ, د/ذ, ر/ز, س/ش, ص/ض, ط/ظ, ع/غ, ف/ق.
- Taa marbuta vs haa: ة and ه at word endings.
- Hamza forms: أ إ آ ا.
- Merged or split words: a lost or extra space.
Quick trick: paste the text into Word or Google Docs with Arabic spell-check on. Red-underlined words are usually OCR errors.
After extraction: what to do with the text
Into Word: paste into a new document, set paragraph direction to right-to-left, and choose a clear Arabic font. Rebuild headings and tables by hand — OCR gives you text, not formatting.
For research: keep the original image next to the extracted text so you can check against it when in doubt.
For translation: paste into your translation tool, but check key terms — OCR errors multiply after translation.
For searchable PDFs: some advanced software adds an invisible text layer over the page image so the PDF becomes searchable while looking the same. WPPR extracts text for copying and saving; it doesn't add a hidden layer to the file.
Special cases
A table in an image
OCR extracts the text but cell order can get mixed, especially in Arabic. Rebuild small tables by hand. For large ones, crop each column into its own image and OCR it separately.
Arabic and English in one image
Choose Arabic + English, and check the order of English words inside Arabic lines afterwards.
Phone screenshots
Screenshots are usually crisp, but small fonts have low resolution. Zoom the content on screen before taking the screenshot.
A whole book
Combine pages into one PDF with JPG to PDF and process it in batches if it's long — OCR uses memory and time per page. And remember that copying and redistributing copyrighted books isn't allowed, even if you extracted the text for yourself.
Worked example: three photos, three results
Take the same printed book page in clear Naskh, captured three ways:
Photo 1: a quick angled phone shot under yellow room light. Lines slope and the bottom of the page is in shadow. Result: top lines readable with scattered errors, bottom lines full of wrong letters, some lines fused. Long proofreading needed.
Photo 2: the phone's document scanner in grayscale, near a window. Straight page, evenly white background. Result: near-complete text, errors limited to some vocalised words and digits. A quick review is enough.
Photo 3: a flatbed scan at 300 DPI grayscale. Result: the highest accuracy, very few errors.
The gap between photos 1 and 2 cost nothing but a minute of care. That's the key lesson: preparation time is far shorter than correction time.
OCR on official documents: don't trust blindly
For contracts, invoices, bank statements or medical reports:
- Never treat extracted text as a substitute for the original in legal or financial matters.
- Check every number: amounts, dates, account and ID numbers.
- Watch stamps and signatures: OCR may "read" them and insert random characters.
- Store the original and extracted text together so you can check one against the other.
Comparing ways to extract Arabic text
| Method | Works with | Accuracy | Privacy | Notes |
|---|---|---|---|---|
| PDF to text | Healthy text PDFs | Excellent (original text) | In your browser | Doesn't work on scans |
| WPPR OCR | Images and scanned PDFs | Good to excellent, depending on image | In your browser | Downloads language data once |
| Google Docs (open PDF or image) | Images and PDFs | Good | Uploaded to Google | Usually loses formatting |
| Google Lens on a phone | Images and live camera | Good for short text | Usually cloud processing | Great for quick copying |
| Paid desktop OCR software | Everything | Excellent | On your computer | Better layout retention, but costly |
Choose based on three factors: file type, how sensitive the content is, and whether you need formatting.
A fast clean-up workflow
For long texts, this sequence saves a lot of proofreading time:
- Paste into Word or Google Docs with Arabic spell-check on.
- Find and replace repeated errors: if the engine keeps reading ى instead of ي at word ends, or adds a space before punctuation, fix them all at once.
- Review red-underlined words one by one.
- Compare numbers with the original by hand.
- Reformat: headings, paragraphs, lists.
- Read the whole thing once quickly — words that are spelled correctly but wrong in context only show up when you read.
When to retype instead
Sometimes typing is faster:
- Very short text — two or three lines are quicker to type than to prepare and proofread.
- Irregular handwriting — you'd correct most words anyway.
- Decorative calligraphy on certificates or invitations.
- Badly damaged images that can't be recaptured.
In between, OCR can still be a useful first draft to correct rather than starting from scratch.
FAQ
Is WPPR's Arabic OCR free?
Yes, completely, with no sign-up. Processing happens on your device; only the language data is downloaded on first use.
Why were my results bad when the image looks clear?
The usual culprits: wrong language selected (English for Arabic text), a decorative font, resolution that looks fine on a small screen but isn't, slight skew, or a coloured background. See the preparation section.
Can OCR read Arabic handwriting?
With limited accuracy. Neat, separated handwriting can give acceptable results; fast cursive handwriting usually isn't reliable enough.
Does OCR keep formatting?
No. It extracts text and line order, but fonts, colours, tables and headings need reformatting.
How long does one page take?
It depends on your device and image size — usually seconds to about a minute. The first run is slower because language data is downloaded.
Can I extract text from a video?
Take a clear screenshot of the frame containing the text, then run OCR on the screenshot.
Bottom line
Arabic OCR works well when you give it what it needs: a straight, sharp image with good contrast, enough resolution and the right language setting. Run the selection test first — if the PDF already has real text, direct extraction is more accurate. And always check numbers, names and letters that differ only by dots.
Start with OCR — extract text and choose "Arabic only" for Arabic documents.