Multi-resolution shape description and clustering of three-dimensional head data

Abstract

The purpose of this study was to develop a 3-D anthropometric sizing method based on a clustering algorithm combined with a multi-resolution description and demonstrate the method with 3-D head data. Wavelet decomposition was adopted to provide flexible descriptions of 3-D shapes on different resolution levels. A block-division technique was then proposed to divide each decomposed 3-D surface into a predefined number of blocks. Afterwards, by using the block-distance metric, each decomposed surface was converted into a block-distance vector. Not only the size information but also the geometric information of the 3-D surfaces are contained in the vector. Finally, k-means clustering was performed on the vectors to segment the sample population into several groups. A total of 378 3-D upper head and face samples were analysed to illustrate the applicability of the method. Clustering was validated by using two measures, size-weighted variances and Clustering Validity Index. K-means clusterings of different variables were compared, including head length and head breadth, the top five principal components from principal component analysis (PCA) on the proposed block distance-based vectors and the block distance-based vectors directly. No obvious difference was found between clustering on the vectors with and without PCA. Lower values of the two measures when clustering on the block distance-based vectors indicated that the proposed block distance-based descriptor is superior to the traditional sizing dimensions of head length and head breadth. Unlike the traditional sizing methods based on key dimensions or derived variables, the method proposed in this study is based on the 3-D shape of the body surface. The proposed block-distance vector reflects not only the overall size but also the local spatial geometric features of a 3-D surface. The new method can be expected to improve the ergonomic design of those products requiring close fitting, such as face shields, goggles and helmets.