INVARIANT THINNING

Abstract

One of the most widely used methods for preprocessing binary images is thinning. The popularity of this method rests on the fact that considerable data reduction is achieved while retaining “essential” properties of the original image. Moreover, topological features, which cannot be verified by a genuinely parallel method (by Minsky and Papert22) are more easily treated in thinned images. For these reasons, many articles on this topic were published in the literature. Most of them are concerned with modifications of existing methods in order to yield “nicer” results out of the thinning process. Also, many results of numerical experiments are available in different publications. The aim of this paper is to show that the quite natural requirement of invariance of the results obtained by thinning leads nearly automatically to a method proposed by the authors in different publications. Moreover, this method is genuinely parallel and well-defined so that it is possible to investigate it theoretically. The practical feasibility of the method is also discussed.