Choosing surface data as the foundation for a theory of object recognition
is not without controversy, particularly given its unsuitability for all
domains and applications.
None the less, it is justified in scenes where
knowledge of the three dimensional scene structure is important,
as discussed in this chapter.