Lecture 17: Active Appearance Models and FORMS

Here T (.) is an intensity template for the object and φ(x) is a spatial warp. The {Bμ(.)} are basis functions of the warps and the {αμ} are the coefficients. The basis function are fixed and the α’s specify different examples of the object. For example T (.) is the image of a face and the φ(.) allow warps to get different shapes of faces. Note that the assumption that the warps can be expressed in terms of basis functions is limited. It is plausible for limited deformations of faces but a very large number of eigenvectors are required to model the movement of lips when speaking (Bregler and Omohundro). It is also not a good way to describe the shape changes of an articulated deformable model – like a person or a cow walking – which are better modeled by parts (e.g., lower leg, upper leg, torso, arms, head) joined at joints, see figure (1) and later part of this lecture.