Split into phases
	Current operation is really only first phase
	Second phase should be guess where adjacent images could be to fill in more control points
Recover from discontiguous image
	Low priority, but maybe if support continue option
Guess overlaps
	Cut down on stitch time, improve accuracty
	Do a statistical test to determine if overlap is constant
	After points match compute deltas to see if we signifigantly truncated and re-calculate if so
	
