Skeletonization by Multiple Scan Order

In this paper, we propose a new skeletonization method based on thinning by Euclidean distance´s ascending order, and use different scan orders to generate candidate results, and synthesize the candidate results to form the final result, the pruning is integrated in the procedure of synthesis. Compared with other existing skeletonzation algorithms, ours is less insensitive to boundary noises and the extracted skeleton is stable. And the only information we need is the shape itself, without any other information such as the number of main branches, etc. And finally our skeleton conforms Blum [1]´s skeleton definition, i.e. the skeletal points are located in the centers of maximal circles which are tangential to boundary points.