Non-standard implementation of DE_CMC and DE_94

[kimikage]

The CMC(l:c) color difference model quantifies the difference from a standard (reference) color.
https://en.wikipedia.org/wiki/Color_difference#CMC_l:c_(1984)

However, the current implementation uses the mean value as the "standard" rather than one of the colordiff inputs.

Colors.jl/src/differences.jl

Lines 328 to 329 in efdb8a5

	# Find the mean value of the inputs to use as the "standard"
	ml, mc = (a.l + b.l)/2, (a.c + b.c)/2

The comment implies that it is intentional rather than a mistake, but the averaging is not the original method. Even if we were to follow the method of using the mean values, the following uses a.l instead of ml, which is not consistent.

Colors.jl/src/differences.jl

Lines 338 to 343 in efdb8a5

	# L* adjustment term
	if (a.l <= 16)
	sl = 0.511
	else
	sl = 0.040975*ml/(1+0.01765*ml)
	end

Although CMC (l:c) is an old model, it appears to still be used in the textile industry, as it is included in the ISO 105-J03 standard.

In practice, there seems to be a method of substituting a reference with an average color of multiple samples of the same product when no reference is available, but that is not the same as averaging inside colordiff.

However, this fix is too drastic to be at the patch level, so I would like to fix this in v0.13.0.

[kimikage]

BTW, DE_94 seems to have a similar problem. 😕
https://en.wikipedia.org/wiki/Color_difference#CIE94

Colors.jl/src/differences.jl

Lines 247 to 253 in 65c4e5a

	# Calculate geometric mean of chroma
	mc = sqrt(a.c*b.c)
	# Lightness, hue, chroma correction terms
	sl = 1
	sc = 1+0.045*mc
	sh = 1+0.015*mc