Automatic classification of female body shape using 3D anthropometric scan data

Area of Science:

• Anthropometry
• Computer Vision
• Apparel Design

Background:

• Conventional apparel sizing fails to capture diverse female body shapes.
• Individual differences in body morphology are crucial for accurate garment fitting.
• Existing sizing systems lack comprehensive representation of morphological diversity.

Purpose of the Study:

• To develop a data-driven framework for classifying female body shapes using 3D body scan data.
• To establish a statistically grounded methodology for body shape classification.
• To provide a foundation for apparel pattern customization and mass-customized garment production.

Main Methods:

• Utilized a dataset of 1,019 Korean women's 3D body scans.
• Extracted key anthropometric ratios (chest-to-hip, hip-to-waist, depth indices).
• Applied analysis of variance (ANOVA) and decision tree algorithms for classification rule derivation, followed by canonical discriminant analysis.

Main Results:

• Developed a taxonomy of nine distinct female body shapes.
• Consolidated shapes into three macro groups: Top Hourglass, Regular Hourglass, and Bottom Hourglass.
• Achieved a robust classification accuracy of 93.6% using canonical discriminant analysis.

Conclusions:

• The proposed methodology offers a statistically sound and reproducible approach to body shape classification.
• The classification framework effectively categorizes female body shapes based on anthropometric data.
• This research provides a practical foundation for improving apparel fit through customized pattern design and mass customization.