Same pixels. Different appearance.: sources & model

Become a color detective. Move the surroundings, uncover the samples, and inspect the actual pixels. Then put a simple explanation to a harder test.

What supports the explanation?

What this model assumes

1. Original digital stimuli, not calibrated replications or clinical tests.
2. The page does not measure gaze, retina, brain activity, physical display luminance or a person’s illusion strength.
3. Personal appearance reports stay in temporary component state. They are excluded from URLs, downloads and analytics.
4. Source JPEGs have compression; the original lossless authored PNG is the equality reference. Social video may alter values.
5. No universal claim that a particular neural mechanism explains every context effect.
6. Static optional stripes; no flashing or staring challenge. Skip a pattern if uncomfortable.
7. Paper texture, printing, lighting and shadows can change the home result.
8. Three quantities: RGB is a digital specification. Luminance is a physical photometric quantity, measured in candelas per square meter (cd/m²). Brightness describes visual appearance. The app measures its image buffer, not the emitted light of your screen or your experience.
9. Lightness has another meaning: In color science, lightness is a relative appearance judgment involving a similarly illuminated area regarded as white. CSS HSL “lightness” is a color-coordinate parameter, not a calibrated measurement of that perceptual quantity.
10. Exact authored stimulus: The equality fixture is an 800 × 360 opaque sRGB raster. Each square contains 12,544 pixels, all RGBA(128,128,128,255). Baseline surrounds are codes 32 and 224. The full neutral mask uses 176. Integer source rectangles avoid edge ambiguity in the exact check.
11. sRGB encoding is nonlinear: For a channel c = code/255, use c/12.92 at c ≤ 0.04045, otherwise ((c + 0.055)/1.055)^2.4. Gray codes 32, 128 and 224 give normalized linear-light values 0.01444384, 0.21586050 and 0.74540421. These are not measured cd/m² or perceived-brightness scores.
12. An explicit, limited hypothesis: Our original stripe targets are 40 × 136 pixels. Short ends contact 80 units of aligned bars; long sides contact 272 units of side bars. The candidate rule subtracts their length-weighted linear-light mean from the target level. It is auditable arithmetic, not a biological simulation.
13. Why White’s effect is useful: Blakeslee, Padmanabhan and McCourt (2016) manipulated aligned and flanking bars independently. The research challenges a universal explanation based only on total light/dark border contact. Our original display is informed by that phenomenon, not a calibrated replication or a guarantee of each viewer’s report.
14. Controlled evidence: Blakeslee and McCourt (2023) kept target luminance at 64 cd/m², independently changed ring width/luminance and remote background, and measured matching judgments. Four observers used a calibrated display. The shown group graph uses mean matching luminance with ±1 standard error of the mean; it is not a percentile chart.
15. Geometry belongs to its protocol: That study’s target radius was 0.5° of visual angle. Ring widths ranged from 0.1° to 24.5°; the largest outer radius was 25°. A browser pixel has no universal angular size. Copying 64 cd/m² into RGB code 64 would confuse two different quantities.
16. Mechanisms need more than a catchy story: Sinha and colleagues (2020) used special experimental displays and observations after sight-restoring treatment to constrain explanations of brightness contrast. Their results argue against treating learned scene interpretation as the only cause. Our page does not reproduce those procedures or settle one exclusive neural pathway.
17. A model is a proposal: Troscianko and Osorio (2023) tested an efficient-coding and spatial-filtering account of color appearance. Their cube figure places an input stimulus above the model’s transformed image. A successful transformation is not a photograph of a thought, and our border arithmetic is not their SBL model.
18. Display and observer limits: Color management, local dimming, nonuniform output, viewing angle, room light and glare can affect the viewing condition. A screenshot checks part of an image pipeline; a physical light measurement requires suitable instrumentation. The same source numbers do not establish identical photons at the eye.
19. A fair null observation: If you do not see an effect, keep that result. There is no score, timing target, age norm or attention diagnosis here. The lesson remains useful through the numeric controls, explicit hypothesis and source evidence.

What has been checked

Analytical reference cases, conservation or transition invariants, finite drawing commands, bounded setup parsing, discovery and route integrity are checked automatically. These checks do not establish anatomical fidelity, learner outcomes or browser/device compatibility. Independent subject review, learner trials, comprehensive accessibility review and browser video encoding checks remain pending.

Each source supports the associated claim. Sources do not certify this implementation or its visuals.

About the cover illustration

Actual Figure 5 from Troscianko and Osorio (2023): rendered cube input above a computational color-appearance model transformation. CC BY 4.0. A research stimulus and model output, not a brain photograph.