If this is the case, why would we have several
different resampling
algorithms with varying levels of quality?
To be rather uncharitable....
The same reason the image resampler ([s] actually, there's like six
complete different implementations used for different cases) in Gimp
is a joke; lots and lots of people who need a resampler, learn just
enough to make one that mostly works, call 'done!' and get back to the
parts of the project they'd rather be doing
'of varying quality' is often a codeword for 'different bugs we don't
fully understand and can't be bothered to track down'. Or worse,
"hey, but it's *really fast!* or 'it always worked fine for me, don't
mess with it'.
(Or worse worse: a grad student learning by doing, overtraining
results for some very narrow use. I'm looking at you again, Gimp)
(Or even worse worse worse: "But _this_ implementation doesn't take my
personal superstitions into account!")
There are some tradeoffs to be made in implementation, but few are
that relevant unless you're dealing with very constrained hardware,
eg, 25MHz ARM with 16kB or working memory.
Cheers,
Monty