What does Colour-Vision Deficiency and Discriminability conclude, in numbers, about “CIELAB and CIEDE2000: Turning "a Visible Difference" into a Number”?
Answer: The two formulas diverge most at middling thresholds. At a threshold of five, CIE76 finds only 0.068 of the pairs colliding while CIEDE2000 finds 0.200, about three times as many; at ten, CIE76 gives 0.264 and CIEDE2000 0.555. At three the two give 0.025 and 0.073. At the very small threshold of one both stay below one per cent, and at the large threshold of twenty they give 0.672 and 0.919. The last line says that 209 pairs are ten or more apart under CIE76 but less than five apart under CIEDE2000: a check based on CIE76 would call these pairs "clearly separable", while under the formula closer to human judgement they are "hard to separate".
Scope of the numbers: Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.025, 0.068, 0.073, 0.200, 0.264, 0.555
| deutan view, 9000 red/green pairs | ||
| threshold | share CIE76 | share CIEDE2000 |
| 1 | 0.004 | 0.009 |
| 2 | 0.013 | 0.033 |
| 3 | 0.025 | 0.073 |
| 5 | 0.068 | 0.200 |
| 8 | 0.173 | 0.421 |
| 10 | 0.264 | 0.555 |
| 15 | 0.489 | 0.791 |
| 20 | 0.672 | 0.919 |
| pairs with CIE76 >= 10 but CIEDE2000 < 5: 209 |
Show the full program
# Chapter 4. p42: CIE76 against CIEDE2000, and the threshold you choose.
# Model: the red/green pairs of p13 (white labels >= 4.5:1) under the
# Machado 2009 deuteranopia simulation; the share of pairs below each
# threshold under both formulas. The thresholds are illustrative.
import math
def lin(c):
v = c / 255.0
if v <= 0.04045:
return v / 12.92
return ((v + 0.055) / 1.055) ** 2.4
def rgb(h):
h = h.lstrip("#")
return tuple(lin(int(h[i:i + 2], 16)) for i in (0, 2, 4))
def lum(c):
return math.fsum((0.2126 * c[0], 0.7152 * c[1], 0.0722 * c[2]))
def contrast(a, b):
hi, lo = sorted((lum(a), lum(b)), reverse=True)
return (hi + 0.05) / (lo + 0.05)
# Machado, Oliveira and Fernandes 2009, Table 1, severity 1.0 (rows).
MACHADO = {
"protan": ((0.152286, 1.052583, -0.204868),
(0.114503, 0.786281, 0.099216),
(-0.003882, -0.048116, 1.051998)),
"deutan": ((0.367322, 0.860646, -0.227968),
(0.280085, 0.672501, 0.047413),
(-0.011820, 0.042940, 0.968881)),
"tritan": ((1.255528, -0.076749, -0.178779),
(-0.078411, 0.930809, 0.147602),
(0.004733, 0.691367, 0.303900)),
}
def sim(c, kind):
# applied to linear sRGB, then clipped to the display range
if kind == "normal":
return tuple(c)
out = []
for row in MACHADO[kind]:
v = math.fsum(m * x for m, x in zip(row, c))
out.append(min(1.0, max(0.0, v)))
return tuple(out)
SRGB_XYZ = ((0.4124, 0.3576, 0.1805),
(0.2126, 0.7152, 0.0722),
(0.0193, 0.1192, 0.9505))
def xyz(c):
return tuple(math.fsum(m * x for m, x in zip(row, c))
for row in SRGB_XYZ)
WHITE = xyz((1.0, 1.0, 1.0))
def _f(t):
d = 6.0 / 29.0
if t > d ** 3:
return t ** (1.0 / 3.0)
return t / (3.0 * d * d) + 4.0 / 29.0
def lab(c):
fx, fy, fz = (_f(v / w) for v, w in zip(xyz(c), WHITE))
return (116.0 * fy - 16.0, 500.0 * (fx - fy), 200.0 * (fy - fz))
def de00(p, q):
# CIEDE2000 with kL = kC = kH = 1 (Sharma, Wu and Dalal 2005)
l1, a1, b1 = p
l2, a2, b2 = q
cm = (math.hypot(a1, b1) + math.hypot(a2, b2)) / 2.0
g = 0.5 * (1.0 - math.sqrt(cm ** 7 / (cm ** 7 + 25.0 ** 7)))
a1p, a2p = (1.0 + g) * a1, (1.0 + g) * a2
c1p, c2p = math.hypot(a1p, b1), math.hypot(a2p, b2)
h1p = math.degrees(math.atan2(b1, a1p)) % 360.0 if c1p else 0.0
h2p = math.degrees(math.atan2(b2, a2p)) % 360.0 if c2p else 0.0
dl, dc = l2 - l1, c2p - c1p
if c1p * c2p == 0.0:
dhp = 0.0
else:
dhp = h2p - h1p
if dhp > 180.0:
dhp -= 360.0
elif dhp < -180.0:
dhp += 360.0
dh = 2.0 * math.sqrt(c1p * c2p) * math.sin(math.radians(dhp / 2.0))
lm, cmp_ = (l1 + l2) / 2.0, (c1p + c2p) / 2.0
if c1p * c2p == 0.0:
hm = h1p + h2p
elif abs(h1p - h2p) <= 180.0:
hm = (h1p + h2p) / 2.0
elif h1p + h2p < 360.0:
hm = (h1p + h2p + 360.0) / 2.0
else:
hm = (h1p + h2p - 360.0) / 2.0
r = math.radians
t = math.fsum((1.0, -0.17 * math.cos(r(hm - 30.0)),
0.24 * math.cos(r(2.0 * hm)),
0.32 * math.cos(r(3.0 * hm + 6.0)),
-0.20 * math.cos(r(4.0 * hm - 63.0))))
dtheta = 30.0 * math.exp(-((hm - 275.0) / 25.0) ** 2)
rc = 2.0 * math.sqrt(cmp_ ** 7 / (cmp_ ** 7 + 25.0 ** 7))
sl = 1.0 + 0.015 * (lm - 50.0) ** 2 / math.sqrt(20.0 + (lm - 50.0) ** 2)
sc = 1.0 + 0.045 * cmp_
sh = 1.0 + 0.015 * cmp_ * t
rt = -math.sin(2.0 * r(dtheta)) * rc
x, y, z = dl / sl, dc / sc, dh / sh
return math.sqrt(math.fsum((x * x, y * y, z * z, rt * y * z)))
def de76(p, q):
return math.sqrt(math.fsum((u - v) ** 2 for u, v in zip(p, q)))
def hsv_code(h, s, v):
# HSV (degrees, 0-1, 0-1) to 8-bit sRGB codes
c = v * s
hp = (h % 360.0) / 60.0
x = c * (1.0 - abs(hp % 2.0 - 1.0))
r, g, b = ((c, x, 0.0), (x, c, 0.0), (0.0, c, x),
(0.0, x, c), (x, 0.0, c), (c, 0.0, x))[int(hp) % 6]
m = v - c
return tuple(int(round((u + m) * 255.0)) for u in (r, g, b))
def hexs(codes):
return "#{:02X}{:02X}{:02X}".format(*codes)
def hue(c):
r, g, b = (u / 255.0 for u in c)
mx, mn = max(r, g, b), min(r, g, b)
if mx == mn:
return -1.0
d = mx - mn
if mx == r:
h = ((g - b) / d) % 6.0
elif mx == g:
h = (b - r) / d + 2.0
else:
h = (r - g) / d + 4.0
return 60.0 * h
W = (1.0, 1.0, 1.0)
def candidates():
reds, greens = [], []
for h in (350, 355, 0, 5, 10):
for s in (0.70, 0.85, 1.00):
for i in range(11):
reds.append(hsv_code(h, s, 0.50 + 0.05 * i))
for h in (100, 110, 120, 130, 140):
for s in (0.70, 0.85, 1.00):
for i in range(15):
greens.append(hsv_code(h, s, 0.30 + 0.05 * i))
keep = []
for group in (reds, greens):
ok = sorted({c for c in group
if contrast(tuple(lin(u) for u in c), W) >= 4.5})
keep.append(ok)
return keep
if __name__ == "__main__":
reds, greens = candidates()
d76, d00 = [], []
for a in reds:
la = lab(sim(tuple(lin(u) for u in a), "deutan"))
for b in greens:
lb = lab(sim(tuple(lin(u) for u in b), "deutan"))
d76.append(de76(la, lb))
d00.append(de00(la, lb))
n = len(d00)
print("deutan view, {} red/green pairs".format(n))
print("threshold share CIE76 share CIEDE2000")
for t in (1, 2, 3, 5, 8, 10, 15, 20):
print("{:9d} {:11.3f} {:15.3f}".format(
t, len([x for x in d76 if x < t]) / n,
len([x for x in d00 if x < t]) / n))
big = len([1 for x, y in zip(d76, d00) if x >= 10 and y < 5])
print("pairs with CIE76 >= 10 but CIEDE2000 < 5: {}".format(big))Single source and citation
Fan, Y. (2026). Simulating Colour-Vision Deficiency and Colour Discriminability: Buttons and Charts as Seen by Red-Green Colour-Deficient Users (Version 1.0.0), Chapter 4: CIELAB and CIEDE2000: Turning "a Visible Difference" into a Number. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-324 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026324,
author = {Fan, Yafu},
title = {Simulating Colour-Vision Deficiency and Colour Discriminability: Buttons and Charts as Seen by Red-Green Colour-Deficient Users},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Eight Per Cent and Four and a Half to One: Why Red and Green Buttons That Pass Contrast Still Collide
- Chapter 2 · Linear Light and Relative Luminance: What WCAG Contrast Computes and What It Leaves Out
- Chapter 3 · The Machado Matrices: Simulating Three Kinds of Dichromacy in Linear sRGB
- Chapter 5 · Safe Colours for Right and Wrong Buttons: a Hue Search and a Luminance-Only Scheme
- Chapter 6 · Chart Palettes: Discriminability Matrices and How Many Series Colour Can Carry
- Chapter 7 · Classical Sources East and West: the Paired Colours of the Kaogong ji and Dalton's 1798 Account
- Chapter 8 · Closing the Loop: a Colour-Vision Check in the Front-End Style Library, the Colour-Vision-Safe Palette Checker and a Pre-Registered Test