Find pointmatches using a KDTree

[acardona]

There are a number of use cases where the approximate position of image tiles is known, and most often, their overlap is very large.

For these cases, comparing SIFT features all to all is a waste of time, and a source of potentially incorrect pointmatches.

Ideally, there would be an extended method with an extra boolean flag to make search be constrained to within a radius. This would turn an n^2 into an n*m, where m << n, reducing computation time enormously.

The one issue I have is: the KDTree implementation is in imglib2, and I understand it's undesirable for mpicbg to depend on imglib2. Short of rolling out a custom solution inside imglib2-algorithm, any suggestions?