Categorization of lower body shapes for adult females based on multiple view analysis

Abstract

Body shapes are generally identified by subjective comparisons of body silhouettes or by calculating ratios of girths. In this study, we developed a reliable and objective categorization method for the lower body shapes of women using principal component (PC) analysis and cluster analysis. A total of 2,488 women aged 18—35 within the 90th percentile of body mass index (34.14) were selected from SizeUSA body scan data. Body measurements chosen for the analysis include buttocks angle and 14 proportional measures of widths, depths, front/back depths, girths, and front/back arcs of key lower body locations. Five PCs were extracted, but only the first three components made a strong contribution to explained variance (PC1: waist to top hip silhouette, PC2: top hip to full hip silhouette, and PC3: buttocks prominence). However, PC4 (abdomen prominence) and PC5 (slope from abdomen point to front hip point) were also critical for representing a distinctive shape. They had a single variable, and therefore each variable was retained as z-score. Three body shape groups were categorized by K-means cluster analysis using three PC scores and two z-scores. In order to provide a simple and intuitive application method for defining a new person’s body shape group, we developed two discriminant functions using raw measurements. Body shape can be classified within our system from body measurements by calculating function scores and comparing them with a bivariate plot of function scores of the body shape groups.