Is it possible to formalize areas such as image processing?
By formalize I mean setup axioms, then derive theorems, and reason about image processing concepts and methods formally. I would say now image processing is pretty informal and ill-defined without strong foundations.
Mumford got interested; http://en.wikipedia.org/wiki/David_Mumford and http://www.amazon.com/Pattern-Theory-Stochastic-Real-World-Mathematics/dp/1568815794/ and many vision pdfs at http://www.dam.brown.edu/people/mumford/vision/papers/
Anyway, theory, such as you request, and practice are currently very far apart. You might look up "persistent homology" for example.