Research map / Vision-language and multimodal / Human and embodied modeling
EnColor: Improving Visual Accessibility with a Deep Encoder-Decoder Image Corrector for Color Vision Deficient Individuals
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes EnColor, a small convolutional encoder-decoder that learns to color-correct images so that their CVD-simulated version matches the original, evaluated on tomato images against linear, contrast, fuzzy and clustering baselines.
Problem
Individuals with color vision deficiencies (CVDs) struggle to distinguish colors in images (e.g., ripe vs unripe fruit), and prior correction methods are simple linear or heuristic contrast enhancers that do not restore the intended colors.
Approach
Train an encoder-decoder image corrector with a differentiable CVD simulator inside the loop, using a pixel-wise MAE between the simulated corrected output and the original image, so the network learns to synthesize colors that CVD viewers perceive as the original.
Key results
- EnColor achieves 16.8% better average MSE than Fuzzy and 29.1% better than Linear across CVD intensities 0.1-1.0 (average MSE 396.67 vs 476.63 and 559.82).
- EnColor has the best average SSIM for MSE-SSIM trade-off only partially: Linear has higher average SSIM (0.942 vs 0.923), while EnColor gives the lowest rate of decrease at high intensities.
- In a 43-participant survey, EnColor received the best average contrast (1.59) and ripeness (1.45) scores on a 1-5 scale where lower is better.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use EnColor instead of Fuzzy correction | Tomato image color correction · Laboro Tomato test set, CVD intensities average | MSE: 476.63 -> 396.67 | Helped |
| Use EnColor instead of Linear correction | Tomato image color correction · Laboro Tomato test set, CVD intensities average | SSIM: 0.942 -> 0.923 | Hurt |
| Use EnColor instead of other tested methods | Tomato image color correction · Human survey, 43 participants | contrast rating: 3.99 -> 1.59 | Helped |
| Use EnColor instead of other tested methods | Tomato image color correction · Human survey, 43 participants | ripeness rating: 3.47 -> 1.45 | Helped |
Surprising findings
- Linear correction achieves higher average SSIM (0.942) than EnColor (0.923), so EnColor is not best on every metric.
- EnColor was not the best at low CVD intensities (0.1-0.2) and Fuzzy outperformed it at the most severe intensities (0.9-1.0) on MSE.
Limitations
- Evaluation focuses mainly on tomato images and a single dataset of about 600 images.
- Survey had only 4 participants with CVD; most participants were simulated viewers without CVD.
- Only deuteranomaly/deuteranopia simulated; the model is not best at low or extreme CVD intensities on MSE.
Experimental setup
Datasets: Laboro Tomato dataset
Benchmarks: Laboro Tomato test set
Metrics: MSE SSIM human survey ratings (contrast, ripeness, color retention)
Baselines: Linear (Petrovic and Fujita 2017) Color Contraster (CC) Fuzzy (Lee and dos Santos 2011) Clustering (Dody et al. 2019) DeepCorrect
Builds on: Machado CVD simulation model DeepCorrect
Closest papers by meaning
- Color Shift Estimation-and-Correction for Image Enhancement (CVPR 2024 · similarity 0.83)
- HVDualformer: Histogram-Vision Dual Transformer for White Balance (AAAI 2025 · similarity 0.83)
- White-Balance First, Adjust Later: Cross-Camera Color Constancy via Vision-Language Evaluation (CVPR 2026 · similarity 0.83)
- ABC-Former: Auxiliary Bimodal Cross-domain Transformer with Interactive Channel Attention for White Balance (CVPR 2025 · similarity 0.82)
- Bright-NeRF: Brightening Neural Radiance Field with Color Restoration from Low-Light RAW Images (AAAI 2025 · similarity 0.82)
- Enabling Region-Specific Control via Lassos in Point-Based Colorization (AAAI 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.