AccessFixer: Enhancing GUI Accessibility for Low Vision Users With R-GCN Model
Low-vision users often struggle to perceive mobile GUIs due to excessively small component sizes, insufficient inter-component spacing, and inadequate color contrast. To address this, we propose the first end-to-end collaborative GUI accessibility repair method. Our approach models spatial, semantic, and stylistic relationships among UI components using a Relational Graph Convolutional Network (R-GCN) and jointly optimizes component size, spacing, and color attributes under design-consistency constraints—ensuring uniform color palettes, equidistant layouts, and scalable adaptations—while preventing new accessibility defects. Evaluated on 30 real-world mobile applications, our method resolves 81.2% of identified accessibility issues on average. We contributed pull requests to 10 open-source projects, with 8 accepted or merged. A user study with low-vision participants confirms statistically significant improvements in interface usability and perceived satisfaction.