Distance-based skeletonization of 3D images

A new algorithm for determining the skeleton of a 3D binary image is presented. A topological approach is taken in which the exterior points of the image are sequentially deleted while preserving the image's topology. The order in which exterior points are processed is important, as different orderings produce different skeletons. In this paper points are processed in ascending distance order as defined by the distance transform of the image. This produces a skeleton that lies along the ridges of the distance transform, and so is positioned centrally within the objects. The algorithm is implemented efficiently using ordered queues such that precalculation of the distance transform is not required, unlike many existing methods. Indeed the distance transform is produced as a side-effect of the algorithm. Results are presented for 3D confocal microscope images.