Application of a Growing Self-Organizing Map to Thinning of Binary Characters with Noise

Igor Farkas, research associate,
Lucius Chudy, research associate
Email: farkas@neuro.savba.sk


Abstract:

We present an improved version of our thinning algorithm, based on growing SOM-like approach, specifically Dynamic Cell Structures (DCS). The algorithm creates an output representation of the pattern in the form of undirected graph possessing the desirable features of a skeleton. The line-like graph topology, which would otherwise be violated by original DCS, is forced by modifications in unit connections' update and winners' search. The algorithm was tested on binary characters and is shown to be robust with respect to boundary noise.


WSOM'97