Hierarchical shape description using skeletons and hierarchical shape discrimination by neural networks
Hiroshi Matsushita, Yoshihiro Mori, Toshio Inui · Systems and Computers in Japan · 1991
Abstract In the shape recognition of a nonrigid three‐dimensional (3‐D) body, there is difficulty in how to describe changes of viewpoint and shape, and how to relate it to an efficient recognition method. This paper proposes a hierarchical description based on multiple‐resolution representation using skeletons as primitives for the shape descriptions. Further, methods for hierarchical learning and discrimination using a neural network model are proposed and a simulation of a global discrimination network has been implemented. The results indicate that features are reflected well by the skeletons, and that flexible, efficient shape discrimination is possible.