Skeletonization and segmentation of binary voxel shapes (PhD thesis)

English summary Skeletons are shape descriptors that are of a lower dimensionality than the shape they describe. They are centered within the shape, capture the shape’s symmetry, and describe the topology and articulation of the shape in a compact manner. These qualities make skeletonization a desirable pre-processing step in a variety of applications ranging from shape segmentation and retrieval, to pose estimation and motion planning. The skeleton was formally defined by Blum in 1967. For 2D shapes, the skeleton consists of curves. Extending the definition to 3D shapes, the skeleton is a 2D structure consisting of manifolds and curves, and is called surface skeleton. Because a 1D structure is more compact and more suitable for some applications such as motion planning, alternative skeleton definitions have been proposed that yield 1D structures, known as curve skeletons.