Graph-Based Shape Abstraction

Abstract

Creating a shape abstraction from a pair of exemplars is an important problem facing researchers in object categorization. Although the problem is receiving growing interest, most approaches assume a one-to-one correspondence among local features, restricting their ability to obtain true abstractions of a shape. In this paper we present a new technique for creating an abstract shape prototype from a pair of exemplars whose features are in many-to-many correspondence. Focusing on the domain of 2-D shape, we represent a silhouette as a medial axis graph, whose nodes correspond to "parts" defined by medial branches and whose edges connect adjacent parts. After establishing a nunty-to-many correspondence between the nodes of two medial axis graphs, we recover the abstracted medial axis graph along with the positional and radial attributes associated with its nodes. The abstracted shapes are evaluated in the context of a recognition task.