Parametric human body modeling for virtual dressing

Abstract

This thesis presents a parameterized 3D human body modeling tool based on shape interpolation. In order to have local control on shape interpolation, human body shape descriptions are anthropometrically segmented. Smooth continuity at the boundaries of individual segments is achieved by weighted interpolation. A 3D editing tool has been developed for segmentation. This tool employs winged edge data structure. This structure allows fast access to vertex neighborhoods. The interpolation weights for vertices are assigned during segmentation. We use decaying functions to assign weights on boundary regions, enabling smooth continuity. Raytracing accelerated by octrees is used for synthesizing images of the resulting models. Anti-aliasing in these images is achieved by super-sampling with jittered patterns. The thesis also introduces a virtual dressing tool. This tool contains a 3D modeling interface and a dressing environment combined with a compositing subsystem, which creates dressed images of virtual replicas by utilizing layers of photographic images of garments and layers of synthesized images of body replicas.