Block Division for 3D Head Shape Clustering

Abstract

In our previous Three Dimensional (3D) anthropometric shape clustering study, block-division technique is adopted. The objective of this study was to examine the sensitivity of clustering results on block-division. Such a block-division technique means to divide each 3D surface into a predefined number of blocks. Then by using a block-distance measure, each surface is converted into a block-distance based vector. Finally, k-means clustering is performed on the vectors to segment a population into several groups. Totally 447 3D head samples have been analyzed in the case study. The influence of block division number on clustering was evaluated by using One-way ANOVA. No significant difference was found between the three block division alternatives. This means the adopted method is robust to block division. © 2009 Springer Berlin Heidelberg.