Fixed topology skeletons

Abstract

In this paper, we present a novel approach to robust skeleton extraction. We use undirected graphs to model connectivity of the skeleton points. The graph topology remains unchanged throughout the skeleton computation, which greatly reduces sensitivity of the skeleton to noise in the shape outline. Furthermore, this representation naturally defines an ordering of the points along the skeleton. The process of skeleton extraction can be formulated as energy minimization in this framework. We provide an iterative, snake-like algorithm for the skeleton estimation using distance transform. Fixed topology skeletons are useful if the global shape of the object is known ahead of time, such as for people silhouettes, hand outlines, medical structures, images of letters and digits. Small changes in the object outline should be either ignored, or detected and analyzed, but they do not change the general structure of the underlying skeleton. Example applications include tracking, object recognition and shape analysis.