The hiererchical clustering index is similar to the k-means index, but it only requires a metrix distance to be defined between different feature points. It has good performance for binary features (such as BRIEF visual features).

Build MEX extension for Octave if matlab is not present (Thanks Jordi Gutiérrez Hermoso for the patch)

Fix to the CMakeLists.txt for matlab mex file, better portability. (Thanks to Alastair Harrison for the suggestion)

Conflicts:
	src/cpp/flann/algorithms/kdtree_index.h