The only formula I found that seems to perform significantly better in
some cases is the one based on the rarely known DIN 99 colorspace.
The DIN 99 color difference has the advantage of being associated to a
Uniform Color Space (UCS) which essentially modifies the L*a*b* coordinates
to make the colorspace look "rounder", or more uniform perceptually.
However, there is a concensus in the scientific community that the CIEDE2000
color difference formula is the best... ...on average.
The disadvantage of CIEDE2000 is that it is not linked to a UCS and is thus
harder to visualize the individual contributions of each parameter.
See for instance this article which details what I summarized in the above
two lines:
For all sorts of reasons, DIN 99 is not well known, particularly in
North-America. The article referenced above looks at ways to enhance the DIN
99 UCS, so there is continuing research on improving what we have now.
In a Symposium last year celebrating the 75th anniversary of the CIE
Standard Observer, it was generally agreed that CIEDE2000 was the current
best choice, and this from people with quite different point of views.
Interestingly, at the same Symposium, the same people were debating on the
validity of the Standard Observer itself! The science is far from perfect,
and there are phenomenas which are scientifically explained better than
others.
You have seen, I have seen, and many others have also noticed, that all
formulas "fail" at some point or another, either when looking at a display
or comparing print samples. However, it is dangerous to globalize a formula
quality by sporadic exposure to "bad" cases, as we tend to forget the other,
usually much more numerous, cases, where we take the formula results and
agree with them without questioning.
