Euclidean skeleton via centre-of-maximal-disc extraction

Abstract A skeletonization algorithm is presented, characterized by two main features: invariance under isometric transformations of the pattern, and recoverability. The algorithm is driven by the Euclidean distance map of the pattern. Invariance under isometric transformations is guaranteed due to the use of the Euclidean distance to compute the distance map; recoverability is guaranteed by the inclusion in the set of the skeletal pixels of the centres of the maximal discs. The skeletonization algorithm includes a beautifying step and a pruning step, which favour the use of the skeleton for shape analysis tasks.