henhaobio
Sign in

bioRxivNeurosciencePreprint

Categorical Color Structure Persists in Color Vision Deficiency

Aimee MartiniD, Doris I. BrauniD, Karl R. GegenfurtneriD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Color vision deficiencies affect color perception and discrimination, but their consequences for the structure and stability of color categories remain less clear. We examined how trichromats, anomalous trichromats, and dichromats assigned 450 physical Munsell color samples to eleven basic color categories and measured the stability of these assignments in a retest two hours later. In each session, participants sorted a randomly mixed set of chips by hand and selected one representative chip for each category. We quantified agreement within and between color-vision groups as well as test-retest reliability within participants. For the 217 chips that all trichromats assigned to the same category, anomalous trichromats used the same category label for 85% of chips, and their category assignments were less stable across sessions as those of trichromats. Within anomalous trichromats poorer red-green discrimination, indexed by higher RG-CAD scores, was associated with lower sorting consistency. Dichromats showed lower agreement with trichromats and anomalous trichromats and greater prototype instability, but their color categories remained systematically organized rather than random. Information-theoretic analyses revealed a graded increase in categorical uncertainty from trichromats through anomalous trichromats to dichromats together with increasing reliance on lightness. Thus, the principal effect of color vision deficiency was not a collapse of color categories, but a graded reduction in the precision with which different shades were assigned to otherwise recognizable categories. In a separate subset of nine anomalous trichromats, EnChroma glasses produced no measurable immediate shift toward trichromatic category structures. Overall, our findings suggest that categorical color representations in anomalous trichromats and dichromats remain remarkably robust despite substantial sensory deficits.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (22 Sept 2026) · CC BY 4.0

Red-green color-deficient observers sorting 450 Munsell chips into eleven basic categories retained structured categories; agreement, retest reliability and prototype stability fell stepwise from trichromats through anomalous trichromats to dichromats.

Earlier work found group-level color naming in red-green color-vision deficiency (CVD) broadly trichromat-like, but rarely tested whether individual assignments hold on retest or track quantitative discrimination thresholds. This study adds same-day retest sorting of 450 Munsell chips (including low-chroma ones), prototype stability, Color Assessment and Diagnosis (CAD) severity scores and information-theoretic measures across three color-vision groups.

Key results

  1. ShownAgreement on color categories depends on color-vision status. Almost all of the gap between groups comes from colors that every trichromat named the same way.

    Within-group κ 0.74, 0.62, 0.49 for trichromats, anomalous trichromats, dichromats (Kruskal-Wallis H = 23.87, p < .001). On 217 unanimous trichromat chips, anomalous trichromats matched 84.6% and dichromats 60.6% of consensus names (held-out trichromats 97.3%); within-group agreement (share giving the modal name) was 100%, 85% and 72% on the 217 consensus chips but converged at 70–72% on the 233 other chips. · Global Agreement Structure; Agreement with trichromat consensus

  2. ShownEach person's consistency between two sessions also falls as the deficiency gets more severe. Dichromats' category prototypes were the least stable.

    Test-retest κ 0.86 (trichromats), 0.74 (anomalous trichromats), 0.70 (dichromats; one dichromat lacked T2); H = 15.2, p < .001. T1 prototypes still in the same category at T2: 100%, 86.36%, 43.18% (H = 20.07, p < .001); mean CIELAB ΔE between T1 and T2 prototypes 7.7, 12.5, 18.0 (H = 8.9, p = .011). · Test-Retest Reliability; Prototypical Color Selection and Stability Over…

  3. SuggestedAmong anomalous trichromats, poorer red-green discrimination went with less consistent sorting. In the five dichromats no reliable relationship was found.

    Higher RG-CAD (red-green CAD) scores went with lower mean pairwise κ with other anomalous trichromats at the first sort (n = 15, Spearman r = −.704, p = .003); dichromats showed no reliable relation (n = 5, r = −.100, p = .873). · Category Consistency and Sensory Precision

  4. ShownColor-deficient observers did not respond at random. Their categories kept their structure but had less sharp boundaries, and lightness carried a growing share of label information relative to hue (hue still carried more in every group).

    Mean per-chip Shannon entropy 0.46, 0.76, 0.84 bits for trichromats, anomalous trichromats, dichromats (H = 104.9, p < .001); normalised mutual information with trichromat consensus 0.72, 0.54, 0.41; value-to-hue mutual-information ratio 0.25, 0.49, 0.70. Dichromat confusions were systematic (green, yellow, brown, grey; purple named blue). · Color-Space Structure, Confusions, and…

  5. SuggestedWearing EnChroma notch-filter glasses for about two hours did not move anomalous trichromats' sorting closer to trichromat categories.

    Nine anomalous trichromats sorted without glasses, then after about two hours wearing them: within-group κ .59 vs .58 (Wilcoxon p = .539); agreement on the 217 consensus chips rose by .02 (p = .24), within the spread seen across separate no-glasses retest sortings. · Enchroma; EnChroma condition

Methods, data and reproducibility

Ten trichromats and 20 red-green color-deficient men hand-sorted 450 physical Munsell chips into eleven labelled basic-category containers and chose a prototype per category, repeating two hours later. Agreement (Cohen's κ, Adjusted Rand Index), test-retest κ, prototype CIELAB shifts, entropy and mutual information were compared across groups and against Color Assessment and Diagnosis (CAD) red-green scores.

  • Trichromat controls (4 men, 6 women, age 24-38 years); trichromacy confirmed with the 24-plate Ishihara test · human · n = 10
  • Anomalous trichromats (13 deuteranomalous, 2 protanomalous; male), diagnosed by Ishihara, Nagel anomaloscope and CAD · human · n = 15
  • Dichromats (2 deuteranopes, 3 protanopes; male); one did not complete the retest · human · n = 5
  • EnChroma subset of anomalous trichromats (7 deuteranomalous, 2 protanomalous): baseline vs glasses sorting; 8 also did a separate no-glasses retest day · human · n = 9
  • Munsell Glossy Edition chips: 320 World Color Survey chips plus 130 desaturated chips · n = 450 chips
Methods & tools
Munsell Glossy Edition chips, World Color Survey (WCS) chip set, Ishihara test (24-plate edition), Nagel anomaloscope, Color Assessment and Diagnosis (CAD) test, EnChroma glasses, RIT Munsell renotation (CIELAB, D65), DaltonLens (Brettel, Viénot and Mollon simulation), Python
Reproducibility gaps
Data and code are promised on Zenodo upon acceptance; no link is given yet. The analyses were run in Python, but no package versions or scripts are listed.
Limitations (6)
  • Many chips in the stimulus set were saturated, which may have made some assignments easier. The authors call for denser sampling of low-chroma and category-boundary regions. · stated by the authors
  • Limiting responses to 11 basic color terms may have constrained answers and hidden finer hue distinctions that CVD observers could make. · stated by the authors
  • There were few dichromats (n = 5, protans and deutans pooled), which limits conclusions about subtypes, and the test of CAD score against consistency in dichromats was limited by sample size. · stated by the authors
  • EnChroma was tested only after about two hours of adaptation, so effects of longer-term wear cannot be ruled out. · stated by the authors
  • The EnChroma comparison had no sham or masked lens, always ran the baseline before the glasses, and had only 9 participants, so expectation and order effects are not controlled. · noted in this summary
  • Groups are small and not matched. The 10 controls (4 men, 6 women) were screened only with Ishihara, while the CVD group was all male with no reported ages, and dichromat retest values rest on 4 people. · noted in this summary

Read this if you study color categorization or color vision deficiency and want test-retest, prototype and information-theoretic measures of how red-green deficits reduce the precision of color categories without abolishing them.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISZeki '18Morimoto '23Tregillus '20Hiramatsu '23Rosenke '20Virtanen '25
66% semantic matchNeuroscience
Semir Zeki, Alexandre Javier, Dimitris Mylonas·bioRxiv·Dec 7, 2018· 2 citations