Valence Normalized Spatial Median for skeletonization and matching

摘要

This paper describes using 3D chain expressions for encoding unit-width curve skeletons and measuring shape dissimilarity. By integrating a robust skeletonization technique with 3D chain coding, the proposed algorithm can compare not only the skeleton topology but also the curvature of skeleton segments. Robustness of the search and potential for retrieving similar objects is studied and shown to be better than graph-based matching. The contributions of our work include generating connected and unit-width skeletons using valence normalization; improving the performance of 3D chain codes for similarity match; and skeleton matching through chain expressions. The advantage of our method lies in its ability to distinguish between relatively similar objects and different poses of similar objects. Experimental results demonstrating the validity of the proposed approach is described.