Reconciling User and Designer Preferences in Adapting Web Pages for People with Low Vision
Yoann Bonavero, Marianne Huchard, Michel Meynard · 2015 · Proceedings of the 12th International Web for All Conference (W4A) · doi:10.1145/2745555.2746647
Summary
This paper addresses the challenge of adapting web pages for people with low vision while preserving the original design intent. Unlike simple approaches that override all colors or apply uniform changes, the authors frame web page adaptation as a multi-objective optimization problem where user accessibility needs (such as minimum contrast between text and background) may conflict with designer preferences (such as keeping colors close to the originals and maintaining uniform background brightness). The approach works at the HTML element level, identifying variation points — properties like text color, background color, and font size — for each element and defining preferences as constraints linking these variables. Three general preferences are defined: uniform background brightness across the page, minimum brightness contrast between text and its background, and substituting original colors with close alternatives. The authors compare two evolutionary algorithms: NSGA-II (Non-dominated Sorting Genetic Algorithm II) and NSGA-III, the latter designed specifically for many-objective optimization problems with more than four objectives. The system was tested on 9 real websites chosen by a co-author with low vision, ranging from simple pages with a dozen elements to complex sites with hundreds of HTML elements and dozens of colors.
Key findings
NSGA-III significantly outperformed NSGA-II across multiple metrics. For simpler preference sets (S1: contrast only), both algorithms found good solutions reliably, but NSGA-III was two to three times faster. For complex preference sets combining all three general preferences (S4: 37 objective functions), NSGA-II could not find any good solution within the 10-second time limit on most websites, while NSGA-III found solutions and achieved approximately 43% satisfaction of objective functions versus about 21% for NSGA-II. Aggregating objective functions — combining related preferences into fewer evaluation functions — improved both algorithms, with NSGA-III plus aggregation yielding the best results overall. The number of simultaneous stimuli (objectives) was the primary difficulty factor: each additional objective function increased computational complexity significantly. On the GoDaddy website example (22 color variables, 14 text colors, 8 background colors), NSGA-III with aggregation found solutions in under 0.5 seconds for simpler preference sets. The approach proved feasible for realistic web pages, demonstrating that automated adaptation can balance user accessibility needs with designer aesthetics in acceptable computation time.
Relevance
This research tackles a practical tension in web accessibility: users with low vision need higher contrast and specific color adaptations, but crude adaptation tools that simply override all styling can destroy the page's visual design, remove meaningful color distinctions, and frustrate both users and designers. The multi-objective optimization approach offers a more nuanced solution that finds the best possible compromise. For web developers, the paper reinforces the importance of using CSS custom properties and semantic color systems that can be systematically adapted — pages with well-structured color schemes are easier for automated tools to adapt meaningfully. The work also highlights that WCAG's brightness contrast requirements, while important, represent just one dimension of the adaptation problem; maintaining color proximity and uniform brightness across the page also matter for usability. As web content grows more visually complex, automated approaches like this become increasingly necessary to make personalized accessibility scalable.
Tags: low vision · web page personalization · color contrast · evolutionary algorithms · content adaptation · visual accessibility · optimization
Standards referenced: WCAG 2.0