Interactive Mesh Deformation that Preserves Visual Saliency

Interactive mesh deformation is a powerful tool to create new mesh models from existing mesh models. When a mesh model is deformed, it is very important to preserve visually distinct features. However, existing deformation methods, which typically maintain discrete mean curvature, fail to preserve such visually distinct features. This is because mean curvature is calculated using just one- or two-ring vertices, although geometric features can be represented in various scales of mesh regions. In our method, the shapes of features are constrained based on visual saliency, which is calculated using a weighted sum of various resolutions of mean curvature values. We introduce a new constraint that can control the stiffness of a feature region based on the value of visual saliency. Such constraints are effective to realize interactive deformation that adequately preserves visually distinct features. We show that our method can interactively deform mesh models while preserving the visual saliency of mesh models.