An Arabic CAPTCHA framework that combines secure user verification with Arabic OCR enhancement — awarded 1st place among all graduation projects.
year
'26
Al
ML
OCR
Problem
Arabic is one of the most widely spoken languages in the world, yet most CAPTCHA systems are built entirely around Latin text — leaving Arabic-language platforms either relying on ineffective English CAPTCHAs or having no robust verification system at all. On top of that, existing OCR models perform poorly on Arabic script due to its cursive nature, right-to-left direction, and limited training data. Arabic users were being underserved on both security and accessibility fronts.
Solution
ArabCaptcha was built to solve both problems simultaneously. The framework generates synthetic Arabic CAPTCHA images specifically designed to challenge bots while remaining readable to humans. To push OCR accuracy further, 8 vision-language and OCR models were evaluated and fine-tuned on 150 Arabic text images across 4 distinct categories. The best-performing models were benchmarked against each other to find the optimal combination of security and recognition quality.
