Accessibility for Color Vision Deficiencies: Challenges and Findings of a Large Scale Study on Paper Figures

Title of the Paper

Accessibility for Color Vision Deficiencies: Challenges and Findings of a Large Scale Study on Paper Figures

Paper Information

• Research Domain: Accessibility studies for color vision deficiencies in visualization images
• Keywords: Accessibility, color vision deficiency, data visualization, crowdsourcing, human-computer interaction, image design, research paper

Research Background and Issues

• Identified Problems or Challenges:

  • Color vision deficiency (CVD) affects individuals' ability to accurately perceive visualization images, increasing the difficulty of information acquisition.
  • Although previous studies have explored accessible data visualization, there is a lack of systematic research on accessibility design specifically for academic research figures.
  • Visualization designers have limited understanding of accessibility design, and many assistive tools and design guidelines remain underutilized.
  • Existing tools and evaluation methods for simulating CVD effects have been rarely applied to large-scale datasets.

• Significance:

  • Approximately 300 million people worldwide suffer from color vision deficiencies, making it difficult for them to interpret color-coded information in images.
  • Data visualization is widely used in scientific communication and information dissemination, and its accessibility is crucial not only for individuals with CVD but also for broader user groups.

• Research Motivation and Related Work:

  • Motivation: Address the gap in large-scale accessibility studies within the visualization domain.
  • Related Work: Previous studies have focused on accessibility in web pages or academic papers in fields like biology and psychology, with limited attention to HCI or visualization research. This study aims to extend these efforts by conducting a large-scale evaluation of visualization images.

Solution

• Methods and Design:

  1. Image Data Selection and Preparation:
     • A sample of 1,710 images was extracted from IEEE Visualization Conference papers, spanning the years 2000 to 2019.
     • The Coblis tool was used to simulate four types of color vision deficiencies (protanomaly, deuteranomaly, tritanomaly, and monochromacy), generating simulated versions for each image.
  2. Label Identification Phase:
     • Four analysts conducted open coding on 210 images to identify accessibility issues and helpful features.
     • A set of labels was developed, covering color-related issues (e.g., indistinguishable colors, low brightness contrast) and general issues (e.g., small font size, low resolution).
  3. Crowdsourced Annotation Task:
     • 200 workers were recruited via Amazon Mechanical Turk to classify and annotate 1,500 images.
     • Annotations included accessibility ratings (accessible, partially accessible, inaccessible), types of issues identified, and positive factors.

• Innovative Aspects:

  • This study sets a new benchmark in exploring accessibility in visualization at a large scale.
  • Combines human vision deficiency simulation with crowdsourcing to analyze panoramic trends and individual differences in large datasets.
  • The newly developed labeling system provides a foundation for designing automated accessibility evaluation tools.

• Key Technologies:

  • Color vision deficiency simulation tools (Coblis), CVD-context data annotation, UMAP data visualization methods, machine learning clustering analysis.

Research Findings

• Specific Results:

  • 60% of images were deemed accessible by most participants, but only 13% of images were completely free of issues.
  • Accessibility significantly decreased under severe CVD conditions (e.g., tritanomaly and monochromacy).
  • The labeling system identified 11 major issue tags (e.g., indistinguishable colors, insufficient brightness) and 6 helpful feature tags (e.g., label text, use of textures).

• Advantages Compared to Existing Solutions:

  • Broader scope, analyzing visualization designs across five years and multiple domains.
  • First to combine data simulation with visual impairment and crowd evaluation, addressing both common and rare CVD cases.
  • Data mining methods revealed individual differences and mainstream trends in issue annotation.

• Experimental and Evaluation Results:

  • Readability issues (e.g., small font size, low image resolution) were more dominant than simple color-related problems.
  • Adding labels and descriptive text was key to improving accessibility, sometimes mitigating image issues.
  • Simpler image designs (e.g., example illustrations) scored higher in accessibility, while complex high-dimensional visualizations (e.g., 3D continuous data) were harder to optimize.

• Limitations and Future Directions:

  • Limitations: Only four types of CVD were simulated, not fully representing the actual experiences of individuals with CVD; no comparative analysis between original and simulated images.
  • Future directions include:
    • Developing personalized accessibility design tools;
    • Exploring accessibility issues beyond color vision deficiencies;
    • Expanding research to more complex scenarios (e.g., interactive visualizations, visualizations across different media);
    • Investigating automated detection tools for evaluating accessibility design.