I haven't really kept up with the latest in image processing, but I've usually seen 3 points as the minimum for image registration. They do it with two. Anyone know of any recent advances in image registration, methods/papers?

Not sure why you need 3 points for 2D image registration. You need four variables here: two to determine the origin (x,y), one each for orientation and scale. You get these variables pinned with 2 points.

Maybe this is a bit simpler, since you know that UP is North.