Image alignment
	Goal:  estimate single (u,v) translation for entire image • Easier subcase:  solvable by pyramid-based Lukas-Kanade