While the application of 3D body shape modeling has been expanding across the apparel industry, including digital fashion, virtual fitting, and personalized clothing recommendation systems, practical service environments face limitations in terms of c...
While the application of 3D body shape modeling has been expanding across the apparel industry, including digital fashion, virtual fitting, and personalized clothing recommendation systems, practical service environments face limitations in terms of cost and operational efficiency regarding the continuous acquisition and utilization of scan-based individual body data. Furthermore, to precisely reflect individual 3D body characteristics through limited anthropometric inputs, it is necessary to select key variables representing body shape variation and to establish a body type classification system based on objective criteria. The purpose of this study was to classify the whole-body types of Korean adult males and females based on body shape indices and anthropometric measurements using data from the 8th Size Korea survey, and to derive a key variable system that can be efficiently applied to 3D body shape modeling.
The data used for analysis consisted of direct measurements and automated 3D scan measurements from the 8th Size Korea survey, integrating cases where both measurements were performed on the same subjects. The final sample comprised 2,165 adults aged 20–59 years, including 1,002 males and 1,163 females. Statistical analysis was performed using SPSS 28.0.
As for the methodology, 71 previous studies related to 3D body shape modeling and body type classification were systematically reviewed to standardize anthropometric items into Size Korea terminology, and a candidate variable set was constructed based on the top 80% of cumulative usage frequency. Subsequently, key input variables were selected based on measurement convenience and representativeness of body shape, and their applicability was examined through descriptive statistics by age group and body shape distributions based on KS K 0050 and KS K 0051 standards. Correlation analysis verified the representativeness of the key variables, and principal component analysis (PCA) was used to derive auxiliary measurements for complementing 3D shape reconstruction, establishing a dual-variable system that distinguishes the analysis stage from the application stage. Based on the final variable set, K-means cluster analysis was conducted to derive body type categories by sex, and canonical discriminant analysis verified classification validity and reproducibility. Furthermore, multiple regression models were established to predict auxiliary measurements, and Random Forest variable importance analysis along with t-SNE and UMAP-based nonlinear dimensionality reduction was employed to cross-validate the statistical classification results.
The results revealed a total of 85 candidate variables through literature-based analysis. Ten key input variables manageable for users in actual application environments were finalized: stature, weight, chest circumference (males) or bust circumference (females), waist circumference, hip circumference, and derived indices including BMI, WHR, WHtR, drop, and lower drop. Correlation analysis showed that these key variables generally represented major anthropometric dimensions (height, trunk volume, central abdominal characteristics, and proportional indices). Through PCA, four auxiliary measurements—abdominal extension circumference, mid-thigh circumference, crotch height, and shoulder breadth—were additionally derived, resulting in a 14-variable system. In K-means clustering, a 10-cluster solution was found to be most suitable for both males and females in terms of distribution balance and convergence stability, with statistically significant differences in anthropometric measurements and body shape indices between clusters. Canonical discriminant analysis confirmed high reclassification accuracy, ensuring the validity and reproducibility of the classification, while auxiliary measurements were proven to be predictable using regression equations based on key variables. Moreover, Random Forest analysis showed that the selected key variables demonstrated consistent importance from a data-driven perspective, and t-SNE/UMAP visualizations confirmed that the cluster structures were clearly separated in the latent space. These findings complementarily verified that the statistical classification results align with the intrinsic distribution characteristics of anthropometric data.
This study is significant in that it presents an optimized key variable system for 3D body shape modeling based on large-scale anthropometric data and establishes an analytical procedure for systematic whole-body classification even with minimal input. The results provide a methodological basis for supporting body shape classification and 3D shape generation using only user-input measurements in environments where 3D scan data are limited, and are expected to have practical utility in digital fashion, virtual fitting, and customized apparel design.