A newer strand of work is described in a http://www.springerlink.com/content/0263v18l50812166/fulltext.pdfpaper from Luuk Spreeuwers [30]. This work provides a higher standard to aim for, having just claimed a 99.0% identification rate for one of the tests on FRGC v2 data. They also address the issue with ICP, which very much resembles the debate over pairwise vs groupwise non-rigid registration. Given the existing framework, plugging in something which emulates the above paper is a task requiring that we resolve all the same common issues, e.g. input preparation, then PCA. Their node-finding method (page 8) is of interest to us too because it is more advanced than some of what we currently have ('ViolaJones', 'icp', 'sphere', 'nearest', and 'nearcentre'). Tables 9-11 do not provides assessment of speed and performance in terms of detection for geodesic distances.