Improving Calibration Time and Accuracy for Situation-Specific Models of Color Differentiation
David R. Flatla, Carl Gutwin · 2011 · The Proceedings of the 13th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2011) · doi:10.1145/2049536.2049572
Summary
This paper presents ICD-2, a significantly improved situation-specific model of human color differentiation designed to make personalised color accessibility tools practical for real-world use. Color vision deficiencies (CVDs) affect people's ability to distinguish colors on digital displays, and recoloring tools can help by transforming problematic color palettes. However, these tools need an accurate model of what colors each individual user can and cannot differentiate. Traditional assumption-based models classify users into broad CVD types (protanopia, deuteranopia, tritanopia) and assume standard severity levels, but this approach fails to capture the full variability of individual color perception, which is affected by genetic factors, acquired conditions (aging, disease, medication), and environmental factors (ambient lighting, display characteristics). Situation-specific models address this by calibrating directly with the user in their current environment, capturing all factors simultaneously. The previous situation-specific model (ICD) used RGB color space, requiring 192 calibration trials across 64 colors — taking over 30 minutes, which was impractical for regular use. ICD-2 replaces RGB with the CIE LUV perceptually uniform color space, which better matches human color vision. In LUV space, the differentiation ability around a color can be described by a discrimination ellipsoid rather than the rectangular box needed in RGB, and only 8 calibration colors (rather than 64) are needed. Each calibration trial uses a gapped circle task where users identify the orientation of a gap in a circle made of colored dots against a background.
Key findings
An empirical study with 14 participants (8 with normal color vision, 6 with CVDs including protans, deutans, and a tritan) compared ICD-2 against the original ICD model. Calibration time was reduced 24-fold: ICD-2 took a mean of 2.17 minutes versus 52.6 minutes for the old model (F=123.46, p<0.001). Overall accuracy improved from 76.1% to 78.7%, a small but significant gain (F=5.13, p<0.05), representing a 2.6% increase. ICD-2 showed a higher false-positive rate (16.02% vs 10.01%, p<0.001) — predicting colors as differentiable when they were not — but a lower false-negative rate (4.75% vs 14.18%, p<0.05). For recoloring applications, false positives are the more concerning error (leaving problematic colors unchanged), while false negatives are less harmful (unnecessarily recoloring already-differentiable colors). The authors note that the higher false-positive rate may be partly due to the step size between calibration colors being too large, causing the discrimination ellipsoid to be smaller than it should be. This could be addressed by using a sigmoid function rather than a step function to model the psychometric transition from non-differentiable to differentiable.
Relevance
This work makes personalised color accessibility genuinely feasible for everyday use. A 2-minute calibration versus 30+ minutes is the difference between a tool people will actually use and one they abandon. For accessibility practitioners, the key takeaway is that color accessibility cannot be fully addressed by assumption-based approaches alone — two people with the same CVD type may have very different color differentiation abilities depending on severity, age, lighting, and display characteristics. The situation-specific approach captures all of these factors simultaneously. The ICD-2 model could enable adaptive recoloring tools that adjust digital content in real time based on a quick personal calibration, benefiting not only people with genetic CVDs but also those with acquired color vision changes from aging, medication, or eye conditions. The model could also inform color palette design tools, helping designers choose colors that are differentiable by the widest range of users in specific viewing conditions.
Tags: color vision deficiency · color blindness · color accessibility · calibration · color differentiation · recoloring · personalization · visual accessibility