In order to improve the underlying framework which aligns the training
set - without taking any or much account of the structure of the
face (e.g. parts to weigh more heavily) - exploration of the ICP
process is undertaken again. There is a lot of work that can be done
on improving it. The first random pairs (random examples) are shown
as the residual, based on ICP with translation and rotation in all
3 axes (see Figure ).
![]()
|