Feasibility study: AI model identifies low-accessibility web pages for color vision deficiency

A new study from arXiv explores the feasibility of automatically filtering web pages that are difficult to access for people with color vision deficiency (CVD). The research, submitted on June 20, 2026, by Ryota Mizutani, Shiori Nakayama, and Masateru Tsunoda, focuses on applying Color Universal Design (CUD) principles to web content. The team built a prediction model that analyzes web pages for low accessibility related to color. In an experiment using 21 web pages, the model achieved a maximum AUC of 0.76, indicating reasonable accuracy in flagging problematic pages. The paper is 4 pages long and includes 3 figures and 3 tables.

The importance of universal design is growing as websites become primary sources of information and services. CVD affects a significant portion of the population, making color-dependent content—like charts, buttons, and highlighted text—challenging to interpret. Current accessibility checks often require manual effort. This automated approach could streamline the process, allowing developers to quickly identify and fix low-accessibility pages. The study demonstrates that machine learning can play a role in web accessibility, though larger datasets and real-world testing are needed to refine the model. Future work may extend the approach to other accessibility dimensions beyond color.