pith. sign in

arxiv: 2606.22095 · v1 · pith:KOF5RN57new · submitted 2026-06-20 · 💻 cs.IR

A feasibility study on filtering low-accessibility web pages considering color vision deficiency

Ryota Mizutani , Shiori Nakayama , Masateru Tsunoda This is my paper

Pith reviewed 2026-06-26 11:06 UTC · model grok-4.3

classification 💻 cs.IR
keywords web accessibilitycolor vision deficiencyuniversal designprediction modelmachine learningaccessibility filteringCUDfeasibility study
0
0 comments X

The pith

A prediction model can identify web pages with low accessibility for color vision deficiency at up to 0.76 AUC.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether machine learning can automatically flag web pages that are hard for people with color vision deficiency to read and navigate. It builds a prediction model from data on just 21 pages and reports a peak AUC of 0.76. This matters because manual accessibility audits are slow, so an automated filter could let designers and search systems surface or improve pages that meet color universal design standards. The work frames the result as evidence that the filtering approach is feasible.

Core claim

Using measurements from 21 web pages, the authors train a prediction model that identifies low-accessibility pages for color vision deficiency with a maximum AUC of 0.76, showing that automatic filtering of such pages is feasible.

What carries the argument

Machine learning prediction model trained to classify web pages by accessibility for color vision deficiency.

If this is right

  • Web pages that fail color universal design can be detected without manual review of every site.
  • Search engines or crawlers could incorporate the filter to prioritize accessible content.
  • Small page samples may be enough to train initial versions of such detectors.
  • Color vision deficiency accessibility can be addressed at scale through automation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If extended, the model could feed into browser extensions that warn users or suggest alternative pages.
  • The same sampling and modeling approach might apply to other accessibility barriers such as screen-reader compatibility.
  • Integration with content-management systems could automatically score and improve new pages before publication.

Load-bearing premise

That the 21 web pages form a sufficient and representative sample for training and evaluating a model that generalizes to real-world low-accessibility detection for color vision deficiency.

What would settle it

Running the identical model on a much larger and more diverse set of web pages and observing that AUC falls well below 0.76 would indicate the reported accuracy does not hold outside the small sample.

read the original abstract

Recently, the importance of universal design has increased. Color universal design (CUD) is one type of universal design that takes people with color vision deficiency (CVD) into consideration. Websites are important media for providing various types of information and functions. Therefore, it is essential to enhance the accessibility of web pages by incorporating CUD principles. The goal of our study is to help improve the accessibility of web pages. Our approach is to automatically filter low-accessibility web pages. To evaluate the feasibility of this approach, we conducted an experiment using 21 web pages. The prediction model identified low-accessibility pages with reasonable accuracy, achieving a maximum AUC of 0.76.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript reports a feasibility study for automatically filtering web pages with low accessibility for users with color vision deficiency (CVD) under color universal design principles. An experiment on 21 web pages trains a prediction model that identifies low-accessibility pages with a maximum AUC of 0.76.

Significance. If the result holds after addressing sample size and methodological transparency, the work could support development of automated accessibility tools in information retrieval and web engineering. The small-scale pilot provides an initial data point on the problem, though its current form does not yet establish practical utility.

major comments (2)
  1. [Abstract] Abstract: The claim that the prediction model 'identified low-accessibility pages with reasonable accuracy, achieving a maximum AUC of 0.76' supplies no information on model type, input features, labeling process, cross-validation, or error analysis. Without these details the reported metric cannot be assessed for overfitting or support for the feasibility conclusion.
  2. [Abstract] Abstract / experimental description: The feasibility conclusion rests on a supervised model trained and evaluated on only 21 web pages. This N is too small to demonstrate generalization, as it provides insufficient statistical power, risks selection bias in page choice, and offers no evidence that the pages span real-world diversity in color palettes, layouts, or contrast distributions.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by a single sentence summarizing the feature set or model family used, even at a high level.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our feasibility study. We address each major comment below and outline the revisions we will make to improve clarity and transparency.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that the prediction model 'identified low-accessibility pages with reasonable accuracy, achieving a maximum AUC of 0.76' supplies no information on model type, input features, labeling process, cross-validation, or error analysis. Without these details the reported metric cannot be assessed for overfitting or support for the feasibility conclusion.

    Authors: We agree that the abstract, as currently written, is too terse to allow readers to evaluate the AUC result. The full manuscript describes the supervised model, color-contrast and CVD-simulation features, expert-based labeling, and cross-validation procedure. To address the concern, we will revise the abstract to include concise statements on model type, feature categories, labeling approach, and evaluation method while remaining within length limits. revision: yes

  2. Referee: [Abstract] Abstract / experimental description: The feasibility conclusion rests on a supervised model trained and evaluated on only 21 web pages. This N is too small to demonstrate generalization, as it provides insufficient statistical power, risks selection bias in page choice, and offers no evidence that the pages span real-world diversity in color palettes, layouts, or contrast distributions.

    Authors: We acknowledge that a sample of 21 pages is small and inherently limits claims about generalization or real-world coverage. The study was designed as an initial feasibility exploration rather than a definitive validation. We will revise the manuscript to explicitly label the work as a pilot study, discuss the risks of selection bias and limited diversity, report the narrow scope of the page sample, and qualify the feasibility conclusion accordingly while pointing to the need for larger-scale follow-up experiments. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical ML performance metric on small sample

full rationale

The paper reports a feasibility experiment training a supervised model on 21 web pages and measuring AUC=0.76 as an empirical performance figure. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. The central claim is a direct experimental outcome rather than a reduction of any quantity to its own inputs by construction, satisfying the self-contained criterion with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities are described in the provided text.

pith-pipeline@v0.9.1-grok · 5645 in / 1006 out tokens · 28627 ms · 2026-06-26T11:06:49.621688+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

10 extracted references

  1. [1]

    Asada, Chromatic Vision Simulator, 2010

    A. Asada, Chromatic Vision Simulator, 2010. https://asada.website /cvsimulator/e/

    2010

  2. [2]

    Computerized simulation of color appearance for dichromats,

    H. Brettel, F. Viénot, and J. Mollon, “Computerized simulation of color appearance for dichromats,” Journal of the Optical Society of America A, vol.14, no.10, pp.2647-2655, 1997

    1997

  3. [3]

    A Study of Color Histogram Based Image Retrieval ,

    R. Chakravarti, and X. Meng, “A Study of Color Histogram Based Image Retrieval ,” Proc. of International Conference on Information Technology (ITNG), pp.1323-1328, 2009

    2009

  4. [4]

    Investigating Color-Blind User-Interface Accessibility via Simulated Interfaces

    A. Jamil, and G. Denes, “Investigating Color-Blind User-Interface Accessibility via Simulated Interfaces ” Computers, vol.13, no.2: 53, 2024

    2024

  5. [5]

    Comparing ten WCAG tools for accessibility evaluation of websites,

    S. Kumar, J. Shree, and P. Biswas, "Comparing ten WCAG tools for accessibility evaluation of websites," Technology and Disability, vol.33, no.3, 2021

    2021

  6. [6]

    A novel heuristic method for quantitative assessment of web accessibility for colorblind ,

    S. Meenakshi, and A. Singla, “A novel heuristic method for quantitative assessment of web accessibility for colorblind ,” Universal Access in the Information Society, vol.23, pp.1957-1977, 2024

    1957

  7. [7]

    Research on the Accessibility of Different Colour Schemes for Web Resources for People with Colour Blindness,

    D. Sajek, O. Korotenko, and T. Kyrychok, "Research on the Accessibility of Different Colour Schemes for Web Resources for People with Colour Blindness," Journal of Imaging, vol.11,no.8, article 268, 2025

    2025

  8. [8]

    Color Universal Design Neural Network for the Color Vision Deficiencies,

    S. Seo, and J. Park, "Color Universal Design Neural Network for the Color Vision Deficiencies," arXiv, 2502.08671, 2025

    arXiv 2025

  9. [9]

    Identifying images in the biology literature that are problematic for people with a color-vision deficiency,

    H. Stevens, C. Winegar, A. Oakley, and S. Piccolo, “Identifying images in the biology literature that are problematic for people with a color-vision deficiency,” eLife, vol.13, RP95524, 2024

    2024

  10. [10]

    Prompt -Driven Color Accessibility Evaluation in Diffusion -based Image Generation Models,

    X. Zhuang, J. Echevarria, and K. Akşit, "Prompt -Driven Color Accessibility Evaluation in Diffusion -based Image Generation Models," arXiv, 2603.09832, 2026. Table 2. AUC of each model combination k-NN naïve Bayes NN random forests ALL 0.62 0.70 0.57 0.50 HIST 0.50 0.69 0.54 0.46 DIST 0.76 0.68 0.62 0.70 RGB 0.55 0.69 0.56 0.59 HSV 0.65 0.63 0.53 0.53 Tab...

    arXiv 2026