PDF version of this entire document
Several attempts have been made (however unsuccessful) to follow an approach of triangle/polygon counting with diffusion distance-based mask dilation, where masks expand and add portions of the faces based on diffusion geometry. Rather than use Euclidean measures, one can just add up the differences, but as a discriminant it fails to work too well. Another approach will be explored instead, with the aim of using diffusion geometry for face recognition (so far the best we got is about 80%).
