UC Irvine

Universal design principles mandate that technologies and services, including mobile apps, should be accessible to all users, regardless of their abilities. However, these principles are often overlooked in development practices. This dissertation addresses the significant gap in mobile app accessibility for users with visual impairments, particularly those relying on Assistive Technologies~(ATs) like screen readers. While existing guidelines and tools aim to improve accessibility, they often fall short in real-world scenarios, especially for dynamic content and interactive elements that require more than static rule-based analysis. This work advances the field of accessibility testing and repair by introducing three key contributions: (1) COALA, a deep learning approach for generating informative labels for unlabeled icons, overcoming biases in previous methods; (2) Groundhog and OverSight, automated tools that detect inconsistencies in app accessibility when using ATs, identifying issues related to both under-access and over-access problems; and (3) TimeStump, a framework for detecting and addressing accessibility challenges caused by dynamic content changes in apps. Through these innovations, this research enhances the accessibility of mobile apps for screen reader users, ensuring a more inclusive experience.