Minimum description length has been used previously to optimise models
of shape and a similar approach can be used in part to improve the
results of AAM formation. However, no analogous consideration has
been applied to intensity values or textures; modes of shape variation
and modes of brightness variation prove to be quite distinct. The
former has not been combined in a particularly helpful way with the
latter and an artificial mixture of the two, which is explicitly controlled
by a coefficient , has been the only implementation attempted.
As was described earlier, in order to generate good models of shape, or more specifically good identification of landmark, warps were applied so that commonalities across the whole set of images get accentuated and a new morphed grid holds all the data.