Skeleton is an important shape descriptor for deformable shape matching, because it integrates both geometrical and topological features of a shape. As the skeletonisation process often generates redundant skeleton branches that may seriously disturb the skeleton matching and cause high computational complexity, skeleton pruning is required to remove the inaccurate or redundant branches while preserving the essential topology of the original skeleton. However, pruning approaches normally require manual intervention to produce visually complete skeletons. As different people may have different perceptions for identifying visually complete skeletons, it is unclear how much the accuracy of skeleton-based shape matching is influenced by human selection. Moreover, it is also unclear how skeleton completeness impacts the accuracy of skeleton-based shapematching. We investigate here these two questions in a structured way. In addition, we present experimental evidence to show that it is possible to do automatic skeleton pruning while maintaining the matching accuracy by estimating the approximate pruning power of each shape.
展开▼
