Extracting Skeletons from Distance Maps

A skeleton is a useful shape descriptor that encodes both boundary and region information for a given object. A widely used approach for skeletonization is to use a distance transform. Although many skeletonization algorithms have been developed, most methods involve the computation of high order derivatives and the evaluation of complex expressions. In this paper, we propose an algorithm that rapidly constructs a coarse graph representation of skeletons without involving the computation of complex high order derivatives. Thus, our method can be used to quickly produce the skeleton prototype of a given image. We also show that smooth skeletons in continuous space can be obtained from the coarse graph using the snake model. Since our method processes a given image as a whole, the presence of multiple objects in an image is automatically detected and the skeletons of those objects are computed simultaneously.