University of Houston

Skinning techniques allow animating high fidelity 3D models with few control parameters by mapping the set of low dimensional control parameters to the high dimensional geometric shape of the 3D model. A manipulation on control parameters will be propagated to the 3D model accordingly, generating transformation and deformation on the model (a pose). One popular skinning technique is skinning from examples, where the skinning model is automatically built from several example poses of the 3D model with different sets of control parameters. If the control parameters are also unknown, the problem can be referred as skinning decomposition, where the examples are decomposed to the skinning model and the corresponding control parameters. This research introduces a linear blend skinning decomposition model that can effectively approximate the skin deformation of nearly articulated models as well as highly deformable models by a sparse, convex bone-vertex weight map and a low number of rigid bone transformations. We also suggest some potential applications of skinning decomposition models on existing performance capture techniques.