Feature-based human model matching for 3D garment transfer

Virtual garment technology including 3D garment modeling and online virtual fitting regains people’s attention with the development of e-commerce platforms. Although current garment CAD system with physics engine can generate fine garment mesh draped on human mesh, this method is still time- and labor-consuming to fulfill personalized online fitting demands due to the great number of garment styles and different sizes. To address this problem and improve virtual fitting efficiency, 3D garment transfer is studied to deform reference garment GR and transpose it from a dressed reference human HR onto a targeted human HT. In this paper, a feature-based approach is introduced to match human models with different shape, pose and mesh topology, in order to drive garment deformation and help achieve plausible targeted human dressing effect. The simplest human matching adopted is one-one mesh vertex mapping for the meshes with same topology. Every mesh vertex of reference garment is projected onto nearest reference human triangle face and locally represented with attachment data. Transposed garment is easily reconstructed via relocating garment vertex above targeted human, but scaled garment needs many successive geometric adjustments if size- or shape-preserving required.