Extended Investigations on Skeleton Graph Matching for Object Recognition

Shape similarity estimation of objects is a key component in many computer vision systems. In order to compare two shapes, salient features of a query and target shape are selected and compared with each other, based on a predefined similarity measure. The challenge is to find a meaningful similarity measure that captures most of the original shape properties. One well performing approach called Path Similarity Skeleton Graph Matching has been introduced by Bai and Latecki. Their idea is to represent and match the objects shape by its interior through geodesic paths between skeleton end nodes. Thus it is enabled to robustly match deformable objects. However, insight knowledge about how a similarity measure works is of great importance to understand the matching procedure. In this paper we experimentally evaluate our reimplementation of the Path Similarity Skeleton Graph Matching Algorithm on three 2D shape databases. Furthermore, we outline in detail the strengths and limitations of the described methods. Additionally, we explain how the limitations of the existing algorithm can be overcome.