Abstract CAPTCHAs are widely used as authentication methods in mobile applications and web-based services to prevent AI bots from gaining unauthorized access. Early CAPTCHAs relied on image-based techniques, distorting text within images to make it difficult for bots to decipher. However, modern AI-powered optical character recognition (OCR) systems can now easily bypass these measures. To address this limitation, image-reasoning CAPTCHAs were introduced. Basic versions of these rely solely on object detection, rendering them vulnerable to advanced AI attacks. In response, we propose a novel image-reasoning CAPTCHA, called IReCAPTCHA, which integrates image understanding, noise mitigation, and mathematical reasoning. Unlike traditional object detection-based CAPTCHAs, our approach challenges users with tasks that require the comprehension of multiple objects, counting, and color recognition. Additionally, we apply noise mitigation techniques to obscure visual information, thereby enhancing resistance to AI-based deciphering. To evaluate the effectiveness of our CAPTCHA, we developed a comprehensive dataset specifically designed to test its robustness against AI attacks. The results are highly promising, demonstrating significantly greater resistance compared to existing image-reasoning CAPTCHAs. These findings suggest that our approach has strong potential to enhance online security by effectively deterring automated access attempts by malicious bots.
Das et al. (Mon,) studied this question.