Person:

Gortler, Steven

This work deals with the problem of practical mesh generation for surface normal approximation. Part of its contribution is in presenting previous work in a unified framework. A new algorithm for surface normal approximation is then introduced which improves upon existing ones in a number of aspects. In particular, it produces better approximations of surfaces both in practice and in the theoretical limit regime. Additionally, it resolves in a simple way some of the problems that previous methods for surface approximation suffered from.

We present a 3d deformation method based on Moving Least Squares that extends the work by Schaefer et al. [Schaefer et al. 2006] to the 3d setting. The user controls the deformation by manipulating a set of point handles. Locally, the deformation takes the form of either a rigid transformation or optionally a similarity transformation, and tends to preserve local features. Our derivation of the closed-form solution is based on singular value decomposition, and is applicable to deformation in arbitrary dimensions, as opposed to the planar case in [Schaefer et al. 2006]. Our prototype implementation allows interactive deformation of meshes of over 100k vertices. For the application of 3d mesh deformation, we further introduce a weighting scheme that determines the influence of point handles on vertices based on approximate mesh geodesics. In practice, the new scheme gives much better deformation results for limbed character models, compared with simple Euclidean distance based weighting. The new weighting scheme can be of use to the traditional skinny based deformation technique as well.

We introduce multi-chart geometry images, a new representation for arbitrary surfaces. It is created by resampling a surface onto a regular 2D grid. Whereas the original scheme of Gu et al. maps the entire surface onto a single square, we use an atlas construction to map the surface piecewise onto charts of arbitrary shape. We demonstrate that this added flexibility reduces parametrization distortion and thus provides greater geometric fidelity, particularly for shapes with long extremities, high genus, or disconnected components. Traditional atlas constructions suffer from discontinuous reconstruction across chart boundaries, which in our context create unacceptable surface cracks. Our solution is a novel zippering algorithm that creates a watertight surface. In addition, we present a new atlas chartification scheme based on clustering optimization.

Complex meshes tend to have intricate, detailed silhouettes. This paper proposes two algorithms for extracting a simpler, approximate silhouette from a high-resolution model. Our methods preserve the important features of the silhouette by using the silhouette of a coarser, simplified mesh as a guide. Our simple silhouettes have significantly fewer edges than the original silhouette, while still preserving its appearance.

Motivated by perceptual principles, we derive a new color space in which the associated metric approximates perceived distances and color displacements capture relationships that are robust to spectral changes in illumination. The resulting color space can be used with existing image processing algorithms with little or no change to the methods.

Cellular patterns are all around us, in masonry, tiling, shingles, and many other materials. Such patterns, especially in architectural settings, are influenced by geometric features of the underlying shape. Bricks turn corners, stones frame windows and doorways, and patterns on disconnected portions of a building align to achieve a particular aesthetic goal. We present a strategy for feature-based cellular texturing, where the resulting texture is derived from both patterns of cells and the geometry to which they are applied. As part of this strategy, we perform texturing operations on features in a well-defined order that simplifies the interdependence between cells of adjacent patterns. Occupancy maps are used to indicate which regions of a feature are already occupied by cells of its neighbors, and which regions remain to be textured. We also introduce the notion of a pattern generator — the cellular texturing analogy of a shader used in local illumination — and show how several can be used together to build complex textures. We present results obtained with an implementation of this strategy and discuss details of some example pattern generators.

We describe how some simple properties of discrete one-forms directly relate to some old and new results concerning the parameterization of 3D mesh data. Our first result is an easy proof of Tutte's celebrated “spring-embedding” theorem for planar graphs, which is widely used for parameterizing meshes with the topology of a disk as a planar embedding with a convex boundary. Our second result generalizes the first, dealing with the case where the mesh contains multiple boundaries, which are free to be non-convex in the embedding. We characterize when it is still possible to achieve an embedding, despite these boundaries being non-convex. The third result is an analogous embedding theorem for meshes with genus 1 (topologically equivalent to the torus). Applications of these results to the parameterization of meshes with disk and toroidal topologies are demonstrated. Extensions to higher genus meshes are discussed.

We present an algorithm to mesh point clouds sampled from a closed manifold surface of genus 1. The method relies on a doubly periodic global parameterization of the point cloud to the plane, so no segmentation of the point cloud is required. Based on some recent techniques for parameterizing higher genus meshes, when some mild conditions on the sampling density are satisfied, the algorithm generates a closed toroidal manifold which interpolates the input and is geometrically similar to the sampled surface.

Linear parameterization of 3D meshes with disk topology is usually performed using the method of barycentric coordinates pioneered by Tutte and Floater. This imposes a convex boundary on the parameterization which can significantly distort the result. Recently, several methods showed how to relax the convex boundary requirement while still using the barycentric coordinates formulation. However, this relaxation can result in other artifacts in the parameterization. In this paper we explore these methods and give a general recipe for "natural" boundary conditions for the family of so-called "three point" barycentric coordinates. We discuss the shortcomings of these methods and show how they may be rectified using an iterative scheme or a carefully crafted "virtual boundary". Finally, we show how these methods adapt easily to solve the problem of constrained parameterization.

We propose a metric for surface parameterization specialized to its signal that can be used to create more efficient, high-quality texture maps. Derived from Taylor expansion of signal error, our metric predicts the signal approximation error - the difference between the original surface signal and its reconstruction from the sampled texture. Unlike previous methods, our metric assumes piecewise-linear reconstruction, and thus makes a good approximation to bilinear reconstruction employed in graphics hardware. We achieve significant savings in texture area for a desired signal accuracy compared to the signal-specialized parameterization metric proposed by Sander et al. in the 2002 Eurographics Workshop on Rendering.