Guided thinning

Abstract

We present a generic improvement for parallelizable iterative thinning algorithms. The improvement achieves a significant better approximation of the medial axis. The values of a previously computed distance map guide the thinning algorithms when to remove a point. We define precise metrics that allows us to make unbiased comparisons among algorithms and to quantify their properties. With the new method, the skeletons are more centered and the reconstruction is more precise, i.e., they characterize the shapes in a more representative way without altering other properties of the underlying thinning algorithm. A surprising result is the fact that some older classical thinning algorithms (and a modification of these as available in the ImageJ software suite) now achieve better skeletons with a smaller number of iterations than more recent algorithms.