We don't have any articles describing photofit but the basic idea is
to 'fit' shape and colour parameters (along with position and lighting
parameters) until we match the face region of the 'photo' as well as
possible. This is then the most likely 3D face that produced the photo
(according to our model).
You can read an article of similar work here:
http://www.mpi-inf.mpg.de/~blanz/html/projects_01.htm
Best wishes,