Structural description theories: is RBC/JIM a general-purpose theory of human entry-level object recognition?

A long-standing problem in structural description theories of object recognition has been the lack of concrete proposals for parts, methods of dividing objects into parts, and relations between parts. Biederman's RBC theory and Hummel and Biederman's JIM model are seminal works because they present one of the first concrete solutions to this very difficult problem: RBC/JIM in turn played a major role in turning object recognition into a burgeoning research area. Here, a review of RBC/JIM as the state-of-the-art structural description theory of recognition is presented. A main conclusion is that there are strong limitations on the scope of objects which RBC/JIM can represent, and hence recognize, because mechanisms for dividing objects into parts and representing parts are not general purpose. Nevertheless, RBC/JIM has promise as a model of geometric object recognition, and there are other directions that may be pursued in the interest of developing a more general-purpose theory.