文档介绍：142 IEEE TRANSACHONS ON WORKS, VOL. 4, NO. 1, JANUARY 1993
Implementation of Parallel Thinning
Algorithms Using Recurrent works
Raghu Krishnapuram and Ling-Fan Chen
Abstruct- In this paper, we investigate the use of recurrent 11, we review the Rosenfeld-Kak (RK) algorithm as well as
works for skeletonization and thinning of binary im- the Wang-Zhang (WZ) algorithm. In Section 111, we show
ages. works are trained to learn a deletion rule and they how these thinning algorithms can be implemented using
iteratively delete object pixels until only the skeleton remains. We
present recurrent work architectures that implement recurrent multistage multilayer works. In particular,
a variety of thinning algorithms such as the Rosenfeld-Kak we present a four-stage work that implements the
(RK) algorithm and the Wang-Zhang (WZ) algorithm. We also RK thinning algorithm, and also a simpler two-stage neural
introduce a modified WZ algorithm which produces work that implements the WZ algorithm. We also suggest
that are intuitively more pleasing. a modification to the WZ algorithm to produce skeletons that
are intuitively more pleasing. In Section IV we present some
I. INTRODUCTION examples of skeletons produced by the works based
on the RK and WZ algorithms. The results of the modified
KELETONS (or stick figures) are pact and powerful
S WZ algorithm ar