Skeletonization of labeled gray-tone images

A gray-tone image including perceptually meaningful elongated regions can be represented by a set of line patterns, the skeleton, consisting of pixels having different gray-values and mostly placed along the central positions of the regions themselves. In this paper, the image is considerd as piecewise constant and a labeled image is created by computing the geodesic distance transformation for each image subset with constant gray-value. A sequential skeletonization process is performed on the labeled image, by employing topology preserving removal operations repeatedly applied to subsets with increasing label value. To obtain a one-pixel-thick skeleton, the topology preservation constraint is disregarded in correspondence with certain configurations in the gray-tone image which would otherwise constitute irreducible patterns.