One crazy_idea I once had was to mine popular paintings and phottographs for aestetic patterns and derive (properties of) an synthetic optimum picture from that.
Not crazy enough -- people have done that :-) I don't have links handy, but there was at least one project which merged pictures of attractive women to create an "average-pretty" face -- it may have been culture specific, that is, they repeated this exercise for pictures of women considered attractive in different countries. As far as I remember, the end result was a predictable meh.
So I'm working for a friend's company at the moment (friend is a small business owner who designs websites and a bit of an entrepreneur) anyway, I've persuaded him that we should research the empirical literature on what makes websites effective (which we've done a lot of now) and to advertise ourselves as being special by reason of doing this (which we're only just starting to do).
One thing that I found absolutely remarkable is how unfilled this space tends to be. Like a lot of things in the broad area of empirical aesthetics it seems like there are a lot of potentially useful results (c.f.http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3485842/ ), but they're simply not being applied- either as points of real practice or of marketing differentiation.
A fascinating gap.