Arabic document OCR is optical character recognition applied to Arabic pages: software that converts an image of text into machine-readable characters. It is harder than OCR on Latin script because letters join and change shape by position, more than half the letters carry dots, vowel marks may or may not be present, text runs right to left while numbers and English run left to right, and documents can use different digit sets and calendars. Judge any Arabic OCR software by testing it on your own documents, field by field, and plan human review for the fields it is unsure of.
Why Arabic document OCR is harder than Latin script
Arabic joins letters, relies on dots, runs right to left and mixes digit sets and calendars, and each of those changes what the software has to do. The table lists each feature, what it is according to the technical sources, and the cases to put in your test set.
| Feature | What it is and why it matters | Put this in your test set |
|---|---|---|
| Cursive joining | Arabic is cursive. Most letters take an isolated, initial, medial or final form depending on their neighbours, and six letters do not join to the next letter. There are no reliable gaps between letters to cut at, and one letter looks different at the start, middle and end of a word. | Several fonts and sizes, including tightly set small print |
| Dots | About 15 of the 28 letters carry one or more dots above or below. A lost dot changes the letter and often the word, so a faint scan or a speck of dust can change the meaning. | Faxed, photocopied and phone-photographed pages |
| Diacritics | Vowel and other marks are encoded as separate marks placed above or below letters. The W3C describes them as unusual in standard text, and surveys note they are often omitted in everyday writing. Where present, they sit above or below letters and must be read or deliberately ignored. Where absent, a word can be ambiguous without context. | Any documents that do carry them |
| Direction | Text is written right to left. Unicode stores characters in logical order, and a display algorithm decides how they appear, so what you see and what is stored need not match visually. | Multi-column pages and tables |
| Digits | Three digit sets exist. European digits 0 to 9, Arabic-Indic digits ٠ to ٩, and an Extended Arabic-Indic variant ۰ to ۹ used for Persian and Urdu, in which several digits look different. All three write the same decimal system, but Unicode encodes them as separate characters. Both Arabic-Indic and European digits are in use across Arabic-speaking regions. Numbers are written with the lowest digit on the right, the same as in English. The Arabic-Indic zero is a single dot. | Invoices, ID numbers and amounts printed in each digit set, and in both on one page |
| Mixed Arabic and English | Mixed text is split into directional runs. A company name, product code or email address inside an Arabic sentence is a separate left-to-right run, so check that the extracted order is correct. | Bilingual forms and lines with codes inside Arabic text |
| Dates | Saudi royal decrees are dated in the Hijri calendar. The decree issuing the PDPL, for example, is dated 9/2/1443H, which is 16 September 2021. The Umm al-Qura calendar is based on astronomical calculation, and the order of day, month and year can differ between documents. | Date fields in both calendars and in different orders |
Arabic document OCR covers printed, scanned, photographed and handwritten text, and these are separate problems. A 2023 survey in Applied Sciences notes that Arabic handwriting varies greatly, and a survey on arXiv treats printed and handwritten recognition as distinct challenges. Image quality matters for every script. The documentation of the open-source Tesseract engine lists resolution, skew, scanning borders and noise among the factors that affect results.
OCR right-to-left text handling has a risk of its own that has nothing to do with letter shapes. An engine can read every character correctly and still return the values in the wrong order, or put a number or a Latin-script code in the wrong place inside an Arabic line. Arabic document OCR therefore needs checking for order as well as for characters, which is why the test set above includes multi-column pages, tables and bilingual lines.
How to test Arabic document OCR on your own documents
Arabic OCR accuracy testing means measuring results on the documents your team actually handles, because no vendor figure replaces that. The steps below work for any engine or service.
- Collect a sample that includes your worst documents. Take every document type, every scanner or source and a share of phone photos. A set of clean PDFs will flatter any Arabic document OCR tool.
- Create the correct answers. Have two people key the correct values for each field independently, then reconcile the differences. This is your ground truth.
- Measure at two levels. Character error rate and word error rate count substitutions, insertions and deletions at character and word level. Then score the fields your process depends on, such as amount, date, ID number and name, because one wrong digit in an amount costs more than a wrong letter in a footer.
- Split results by feature. Report digits, dates, mixed-language lines, stamps, small print and handwriting separately. An overall average hides the failures that matter most in Arabic document OCR.
- Test search, not only appearance. Pick known words from the source documents and search for them in the extracted text. Text that looks right on screen but cannot be found is a defect.
- Set review thresholds. Decide which fields a person must check, and route low-confidence readings to them. Ask the vendor how confidence scores are produced and test whether they track real errors.
- Repeat when something changes. A new form version, scanner or engine release can shift Arabic document OCR results.
Handle the output with care
The output of Arabic document OCR is only as useful as the checks around it. Keep the original image, the extracted value and its confidence together. Record a digit or date exactly as read before converting it, and note which calendar and digit set it used. Validate against rules where you can, such as ID formats and amounts that must add up, so that Arabic document OCR errors are caught before they reach another system. Convert Hijri and Gregorian dates with a routine built for the Umm al-Qura calendar, because a simple rule of thumb can be a day out. What happens to the data afterwards is often a rule-based workflow, which is where RPA often takes over.
Personal data and where processing happens
Scanned contracts, ID copies and CVs contain personal data, so the Personal Data Protection Law applies to Arabic document OCR as it does to any other processing. If a service runs outside the Kingdom, check whether sending your scans to it counts as a transfer under the PDPL’s transfer rules. Our guide to data residency in Saudi Arabia covers hosting questions. This is general information, not legal advice. Confirm with a qualified adviser or SDAIA.
For organisations that want Arabic document OCR inside a wider workflow, WASNA Digital Solutions describes intelligent document processing for invoices, contracts and government letters, including Arabic and bilingual scans, on its AI and smart automation page, with accuracy tested on the client’s own documents before rollout.
Bring a sample of your Arabic documents
Send a description of the document types, the scan sources and the fields you need. A WASNA Digital Solutions consultant will propose how to test Arabic document OCR on them before any rollout is agreed.
Sources
- The Unicode Standard, Version 16.0, Chapter 9, Middle East I: Arabic. Unicode Consortium. unicode.org. Accessed 2 October 2026.
- UAX #9: Unicode Bidirectional Algorithm. Unicode Consortium. unicode.org. Accessed 2 October 2026.
- Arabic Layout Requirements. W3C. w3.org. Accessed 2 October 2026.
- Faizullah, S., Ayub, M.S., Hussain, S. and Khan, M.A. A Survey of OCR in Arabic Language: Applications, Techniques, and Challenges. Applied Sciences 13(7), 4584, 2023. MDPI. mdpi.com. Accessed 2 October 2026.
- Kasem, M.S., Mahmoud, M. and Kang, H.-S. Advancements and Challenges in Arabic Optical Character Recognition: A Comprehensive Survey. arXiv:2312.11812. arxiv.org. Accessed 2 October 2026.
- Improving the quality of the output. Tesseract OCR documentation. tesseract-ocr.github.io. Accessed 2 October 2026.
- Eastern Arabic numerals. Wikipedia. en.wikipedia.org. Accessed 2 October 2026.
- Umm al-Qura calendar. Wikipedia. en.wikipedia.org. Accessed 2 October 2026.
- Saudi Arabia’s Personal Data Protection Law: what you need to know, November 2023. Herbert Smith Freehills Kramer. hsfkramer.com. Accessed 2 October 2026.
- Guide to the Saudi Personal Data Protection Law (for controllers and processors). Saudi Data and AI Authority (SDAIA). dgp.sdaia.gov.sa. Accessed 2 October 2026.
