The author shows how a representation for visual shape can exploit knowledge of the ways in which shapes are related by deformation of their geometries. The approach is based not on generic rules or principles for interpreting shape deformation in general, but instead relies on careful analysis of shapes occurring in a specific domain. The representation uses a relatively large vocabulary of simple geometric descriptors, each of which resembles a template that deforms as a mechanical linkage. Because these descriptors are designed to name significant aspects of spatial deformation known to structure the target domain, the representation provides a rich vocabulary with which to perform shape classification, comparison, and other visual tasks involving subtle distinctions about objects' forms. The approach is illustrated for the two-dimensional shape world of fish dorsal fins.
展开▼