See an image through the two published models of colour vision deficiency, Machado 2009 and Vienot 1999, computed in linear light where those models are defined, with the colour pairs that stop being distinguishable listed by name.
Related reading
Initializing in your browser…
A tone curve drawn with its level count and clipping reported, per-range gains in stops of light whose three weights sum to exactly one, and a white balance that is a Bradford adaptation on a published locus. The neutral setting is proved to change nothing.
Extract a palette whose every colour is one the image actually contains, with the share of the picture each one covers. Median cut, k-means in CIE Lab or plain frequency, each named.
Whole-pixel crops with the rectangle typed or dragged, checked by decoding the output and comparing it with the pixels that went in. Quarter turns and mirrors are exact permutations, so nothing is clipped.
A two-series chart whose only difference between the series is hue: a common red and a common olive green.
Input
chart.png · Machado 2009 · Deuteranopia
Colours that stop being different
#DC3545 and #788C3C normal vision dE2000 53.8 (unmistakably different) deuteranopia dE2000 0.4 (the same colour)
A dE2000 of 53.8 is about as different as two colours of similar lightness get; 0.4 is below the threshold at which anyone can see a difference at all. So the two series are the same colour to roughly 1 in 12 men, and the legend is unreadable. The fix is not a different hue, it is a second channel: a pattern, a label or a difference in lightness.
About 8 percent of men and 0.4 percent of women of European descent have a red-green colour vision deficiency (Birch, JOSA A 29(3), 2012), and deuteranomaly alone accounts for roughly 5 of those 8 points. If a design carries meaning in a hue difference alone, a large group of people cannot read it. Upload an image and see it through either of the two published models, with the pairs of colours that stop being distinguishable listed by name and by how far apart they were.
Find the two colours a layout relies on that a large group of people cannot tell apart, by name and by number.
Check whether a chart carries its meaning in hue alone, which is the usual failure.
Confirm brand colours stay distinguishable, or find out which pair does not.
A pair that collapses needs a second channel, that is, a shape, a label, a pattern or a difference in lightness. Adjusting the hue rarely helps.
Both published models of colour vision deficiency operate on LINEAR LIGHT, and that is the sentence most simulators get wrong. The bytes in a PNG are gamma-encoded sRGB; multiplying them by a matrix defined for linear RGB is wrong before you ask which matrix. Measured over 33 swatches, the difference reaches 38.6 dE2000 for protanopia and 55.5 for deuteranopia, and it is not subtle: pure green under deuteranopia is a bright yellow (214, 214, 46) when the model is applied where it is defined, and a dark muddy brown (96, 76, 76) when it is applied to the stored values. A person with deuteranopia does not see green as dark. Everything here undoes the sRGB transfer function first (IEC 61966-2-1), applies the matrix, and encodes the result again.
Two models are offered and each is named with its paper. Vienot, Brettel and Mollon (Color Research and Application 24(4), 1999) convert to LMS cone space and project onto the plane spanned by the two remaining cone responses; it is defined for dichromacy, the complete absence of a cone class, and publishes no severity. Machado, Oliveira and Fernandes (IEEE TVCG 15(6), 2009) derive the shift in cone spectral sensitivity that an anomalous trichromacy is, and publish eleven matrices per deficiency from severity 0 to 100 percent in steps of 10; this interpolates between the two either side. Machado is the one to use for a partial severity, because an anomalous trichromacy is a continuum rather than a switch, and because most colour vision deficiency is anomalous rather than dichromatic. Where a condition is a complete absence, the severity control is disabled and says why. Achromatopsia is covered by neither paper and is shown as relative luminance in linear light, which the page states rather than implying a model exists.
The picture is illustrative. The measurement is the pair list. The tool counts the colours the image is actually made of, groups them with median cut (Heckbert, SIGGRAPH 1982) so that every colour it reports is one the image really contains, and compares every pair before and after with CIE dE2000. A pair that starts at least 10 apart, which is unambiguously different, and ends under 5, which is the same colour at a glance, is listed with both numbers. Those two thresholds are choices and the page says so. On a two-series chart drawn in a common red and a common olive green, the two are 53.8 dE2000 apart to normal vision and 0.4 apart under deuteranopia: the same colour, and the legend is unreadable.
Alpha is left exactly as it was, because a simulation is about which cones respond and a transparent pixel is not more or less transparent to anyone. Everything runs in the browser and the image is never uploaded.
Machado 2009 if you want a severity, which is most of the time: anomalous trichromacy is a continuum and severity 60 percent looks quite different from 100 percent. Vienot 1999 for the complete absence of a cone class, which is what that paper covers.
Most apply their matrix to the stored, gamma-encoded bytes rather than to linear light, which is not what either paper defines. The difference reaches 55 dE2000 and it changes the answer qualitatively: applied to the stored values, pure green under deuteranopia comes out a dark brown, where the model says it is a bright yellow.
Deuteranomaly, at a moderate severity. It is by far the most common form, about 5 percent of men, and it is an anomaly rather than a total absence, so the dichromat simulations most tools show are a worst case rather than a typical one.
It reproduces the published models to within one level of 255, checked against a second transcription of the same tables. What the models themselves capture is the cone response, and nothing about the brain's adaptation, a lifetime of learning which things are which colour, or the light you are reading in. Someone with deuteranomaly is usually far better at telling these colours apart than the picture suggests, which is why the pair list matters more than the picture.
Neither published model covers it. The rendering here is relative luminance in linear light, which is what remains when no cone class discriminates, and the page says so rather than implying a model.
No. Decoding, simulation and analysis all happen in your browser.
Images are decoded, edited, and exported entirely inside this browser tab. No originals, exports, or metadata are uploaded.