Enhancing User Accessibility in CAPTCHA Systems Without Compromising Security

Abstract

CAPTCHA is commonly employed to secure online services from bot attacks; yet, conventional CAPTCHA designs can pose accessibility issues to visually, audibly, and cognitively impaired users. This paper presents an accessible CAPTCHA framework that improves the user experience for people with visual, auditory and cognitive impairments while remaining secure against the latest AI-powered bot attacks. The proposed system is unique in the way it combines a multi-modal approach for generating CAPTCHAs, behavioral biometrics-based verification, and an adaptive adjustment of challenge difficulty into a single authentication system. The multi-modal approach adaptively chooses image, audio, or logic-based CAPTCHAs according to user preferences, enhancing user experience for different user populations. Mouse dynamics, keystroke patterns and response times are employed as behavioral biometrics using a lightweight machine learning model to improve human-bot distinction. Further, a dynamic security mechanism dynamically controls CAPTCHA difficulty based on risk scores, enhancing user experience while maintaining security. The proposed system is evaluated on a dataset of 10,000 CAPTCHA challenges with 500 human users and 1,000 bot simulations and shows substantial improvements. The new system delivers an accuracy of 96.8%, precision of 95.4%, recall of 96.2% and F1-score of 95.8%, with a false acceptance rate (FAR) of 1.9% and false rejection rate (FRR) of 2.3%. More importantly, the usability aspect of the system also increases substantially, going from 68.4% to 92.5%. This shows how effective the system is at finding a balance between security and accessibility, thus making it a feasible alternative for future web authentication systems.