3D Shape Modeling Using High Level Descriptors

The goal of this Ph.D. project is to investigate and improve the methods for describing the surface of 3D objects, with focus on modeling geometric texture on surfaces. Surface modeling being a large field of research, the work done during this project concentrated around a few smaller areas corresponding to the research papers presented here. One of those areas is formulating surface priors by utilizing local surface properties. A well defined prior can, in a Bayesian framework, assist many common task in geometry processing, like denoising, object recovery, object matching and classification. Some of the priors described here are defined on the main entities of the triangular mesh, vertices, edges and faces. Other priors are defined on small planar patches, denoted surfels. Another area of research deals with textures which cannot be described by height fields, for example biological features like thorns, bark and scales. Presented here is a simple method for easy modeling, transferring and editing that kind of texture. The method is an extension of the height-field texture, but incorporates an additional tilt of the height field. Related to modeling non-heightfield textures, a part of my work involved developing feature-aware resizing of models with complex surfaces consisting of underlying shape and a distinctive texture detail. The aim was to deform an object while preserving the shape and size of the features.