Resolution Influence on 3D Anthropometric Data Clustering for Fitting Design

Sizing based on 3D anthropometric data may lead to significant improvement in fitting comfort of wearing products. However, the required computational load is a common problem in 3D data processing. In a previous study, wavelet analysis was adopted to establish a multi-resolution description of 3D anthropometric data to reduce computational load and modeling complexity. K-means clustering was subsequently performed on the decomposed 3D samples. This study further examines the influence of decomposition level on clustering results. As a case study, 378 face samples, 447 head samples and 432 upper head samples were analyzed. Cluster membership variation on five different resolution levels was examined by using Cluster Membership Accuracy Rate (CMAR), which denotes the clustering consistency on the decomposed levels compared with the clustering results on the original data sets. For the face data sets, the CMAR values on the five decomposition levels are 100, 99.21, 97.88, 93.92 and 93.39%, respectively; for upper heads, the CMAR values are 99.3, 99.1, 98.4, 92.1 and 84.3%, respectively; while for whole heads, the CMAR values are 99.3, 98.2, 95.1, 85.5 and 77.9%, respectively. These results indicate that clustering on the third decomposition level is proper for face and head scans in reducing computational load while maintaining at least 95% clustering accuracy.