This is very clever underlying technology, licensed from Singular Inversions of Vancouver. Basically, these guys did several hundred face scans, acquiring detailed texture maps at the same time. . .and then did Principal Components Analysis on both the geometry and texture variation-- essentially identifying the "dimensions" of variation in "FaceSpace".

This was based --at least in part-- on theoretical work by Thomas Vetter, which you can see at: http://graphics.informatik.uni-freiburg.de/CAFE.html

At SI's site (www.facegen.com), you can see their own "FaceGen" software, of which the Face Room appears to be a subset. The biggest problem with FaceGen for generating new heads is that they don't allow distribution of the new ones, and they have nothing like the .CR2 file format for "morph districution". . .a restriction which seems not to be in place with Face Room content.

I apologize in advance if this is a) boring or b) the wrong forum for this post. . .