ABSTRACT
This chapter considers how a computational theory of face perception may eventually allow people to answer the question: what are the "structural codes" for face recognition?. An important aspect of D. Marr's framework was his specification of the different levels at which theoretical statements can be made. The "top" level of theory is a theory of the computation which the visual system must perform in order to solve some particular visual task. The goal of early visual processing, according to Marr, is to describe the surface layout in the viewed scene. What might be the primitives of a representation of a face? Marr and H. K. Nishihara argue for volumetric primitives in the description of certain classes of natural shapes which have the approximate shapes of generalised cones.
