Tandem
Workshop
on Geometry, Numerics
and Visualization
Overview of
Talks
Detection of Shape
Features by Skeletonization
Shape
matching
between triangulated surfaces requires detection and identification of
mutually
corresponding features on shapes. In order to localize such
features, a
set of discrete
curvature operators can be applied. First we translate various
types/notions of
curvatures
known from differential geometry to discrete, piecewise linear
surfaces,
and discuss some of the problems arising from generalizing curvature
notions to
triangulated
surfaces.
Thresholding the curvature field defines
which nodes
of the surface
potentially are shape features. The final step comprises the skeletonization of the feature
field, resulting in a set of feature lines.
Interior Point
Methods: An Introduction
proposed his algorithm for linear programming. In the years since then,
algorithms and software for linear programming have become
quite sophisticated, while extensions to more general classes of
problems,
such as quadratic programming, nonconvex
and
nonlinear problems
have reached varying levels of maturity. In this talk we will introduce
interior-point methods for linear programming.
This talk is in preparation for the following talk from Martin Weiser:
Adaptive Multilevel Methods for Optimal Control Problems.
Biomechanical Modeling of Deformable Soft Tissues
is essential to many computer assisted medical applications.
Since soft tissues are non-rigid, any interaction with
biological structures causes their deformation. Thus, modeling
tissue deformations under the impact of external forces is
of general importance.
In this talk, a brief overview over biomechanical modeling of
deformable biological tissues with emphasis on the quasi-static
FE-approach for the long term soft tissue prediction in
the craniofacial surgery planning is presented.
Creating 3D
Geometrical Models from 3D
Image Data
3D
imaging
methods are widely used in various areas like medicine,
biology, material sciences and earth sciences. In this talk we focus
on biology and medicine, where 3D image data form the basis of
morphological and functional analyses, as well as of image-guided
diagnosis and computer assisted therapy.
A central task in such applications is the creation of 3D models
that faithfully represent the objects depicted in the image data.
Some problems of this task will be sketched. A set of practically
useful methods will be presented which enables researchers to segment
biomedical images and to reconstruct smooth 3D geometrical models.
This core technology provides the basis of a wealth of new quantitative
methods -- in the long term turning many biomedical fields into more
quantitative areas. Examples from various areas like medical treatment
and surgery planning, neuroanatomy and
gene
expression analysis will
be presented.
Denoising and enhencement
of surface features
Noise
is an omnipresent artifact in 2d and 3d meshes due to resolution
problems in
mesh acquisition process. For example, meshes extracted from
image data or
supplied by laser scanning devices often carry high-frequency
noise in the
position of the vertices. Many filtering techniques have been
suggested in
recent years, among them Laplace smoothing
is the
most prominent
example. We present a method for anisotropic denoising
that concentrates on
the preservation and enhencement
of linear and curved surface features. The
method havily
relies on explicit measures for curvature of discrete surfaces.
Reorientation of a
liquid surface -
solving the Navier Stokes equations in a
time-dependent domain
surfaces under various acceleration conditions is of high interest for
the construction of space vehicles using liquid propulsion. This
behavior may be numerically simulated by solving the incompressible
Navier-Stokes equations.
Since the free surface is moving, the computional
domain has to change
as well with the time during the simulation. This talk shall
present
and discuss one technique to solve PDEs in
time-dependent domains.
one-sided contact problem, which
is also known as the
Signorini
problem. Its
aim is to
simulate the mechanical
behaviour of an
elastic body in
the presence of a rigid
obstacle. We explain the standard discretizations
used to tackle
this problem
and, if time permits, introduce
the audience
to fast multigrid solvers for the
Signorini
problem. For Part II see Oliver
Sander.
3d Statistical
Shape Models for Image
Segmentation
medical diagnosis and therapy planning. Since manual image
segmentation is rather time consuming, automatic and robust
segmentation techniques are of great practical importance.
For this, a-priori knowledge about the anatomy has to be
considered. Statistical shape models have proven
to be effective yet difficult to construct in 3D,
because of the problem of identifying corresponding
points that are densely distributed on two shapes.
In this talk the mathematical problems involved with
shape matching are sketched. A solution for matching
geometrical objects of arbitrary topology based on their
representation as triangular meshes is presented. It will
then be shown how to construct statistical shape models
and how to use them in medical image segmentation.
Anisotropic fairing
of point sets
The use of point sets
instead of
meshes became more popular during the
last years. We present a
new method for anisotropic fairing of a point
sampled surface
using an
anisotropic geometric mean curvature flow.
The main advantage of our
approach
is that the evolution removes noise
from a point
set while it
detects and enhances geometric features of
the surface
such as edges
and corners. We derive a shape operator,
principal curvature properties
of a point set, and an anisotropic
Laplacian
of the surface. This anisotropic Laplacian
reflects curvature
properties which can
be
understood as the point set analogue of
Taubin's
curvature-tensor for polyhedral surfaces. We
combine these
discrete tools
with techniques
from geometric diffusion and image
processing. Several
applications
demonstrate the efficiency and
accuracy of our
method.
On the
approximation of geometric
invariants of a smooth surface
shall study the relationship between the geometry of S and the
geometry of T. In particular, we shall compare the area and the
curvatures of S and T. Moreover, we shall explain why these kind
of
problems appear in different fields of science, and we
shall deal with an example in structural geology.
Simulation of
particle systems
In this talk
I will present an algorithm to simulate a system
of particles
or rigid objects interacting on each other. Given
such a set of rigid
objects. Next define some
forces applied on them. They could for example
attract themselves. The
simulation should now calculate what happens
to the objects in real time. With this algorithm it is possible to
visualize
physical processes, like a
spinning gyro, as mathematical ones, like the
moving of a flexible Steffen
Sphere.
Konrad Polthier
(TU)
Conformal
Maps and
Non-conforming Meshes
In
non-conforming simplicial meshes the
continuity
of adjacent triangles is required at edge midpoints only rather
than along whole edges. This relaxation provides additional freedom
which allows us to successfully study some optimization problems for
regular triangle meshes including minimization of the conformal energy.
Recognition of 2D
Vector Field
Singularities Using a Discrete Hodge Decomposition
Singularities of vector
fields are
among the most important features of
flows. Vortices can influence the flying abilities of aircrafts, higher
order singularities often appear in magnetic fields. One approach for
the detection and analyzation of 2d
singularities is
to decompose the
vector field into a rotation free, a divergence free and a harmonic
component. The potential respectively co-potential
of the
first two
components offer an easy way to get some insight into the vector field
structures.
In this talk a discrete Hodge method is presented and an overview of
vector field singularities and their decomposition and detection is
given.
Smoothing
subdivision of discrete
surfaces using the Rivara algorithm
As
the Rivara algorithm itself does not
smooth surfaces,
it is combined
with the Butterfly subdivision scheme to a smoothing local subdivision
algorithm. The Butterfly subdivison scheme
works
usually with a 4-to-1
split of triangles. A great disadvantage of the 4-to-1 split is the
need
of temporary refinements to hold the conformity of the triangulaiton;
these must be removed again, if the subdivision process shall be
continued later. In contrast a Rivara
refined
triangulation is
conformized in a way, that does not require
to distinguish between
triangles refined to improve the geometrical properties of the surface
and triangles refined to gain conformity. Using Rivara
refinement and
Butterfly scheme together combines the good smoothing properties of the
Butterfly scheme with the very pleasant triangulation handling of the
Rivara algorithm. Thus the Rivara
refinement becomes an interesting
algorithm for a local subdivision not only for two-dimensional
triangulations, for what it is mainly used up to now, but as well for
triangulaitons of surfaces in
higher-dimensional
space.
Mathematische
Modellierung von Zweikörperkontakt (II)
Part II treats the behaviour of two
elastic bodies
in contact
with each
other. This two-body contact problem
is
considerably less well
understood. We discuss several ways
of
formalizing and discretizing
the contact conditions and
explain some of
the problems
encountered on the way.
We will further show
where two-body
contact fits into the framework
of
human-gait simulation.
Alfred
Schmidt
(Uni
Adaptive
finite element
methods for phase transition problems
During
solidification of an undercooled melt,
anisotropic interface
effects lead to a geometrical law of motion for the rapidly moving
front between liquid and solid material, the Gibbs-Thomson equation
which couples velocity, anisotropic curvature and temperature.
The interface motion is coupled to heat conduction in the solid and
liquid material.
Different models can be used for the moving interface:
- a parametric surface representing a sharp interface,
- a level set approach,
- a diffuse interface modeled by a phase variable.
All those models lead to degenerate parabolic partial differential
equations for the interface motion.
After introducing the models, we present adaptive finite element
methods for sharp and diffuse interfaces.
The talk presents joint work with E. Bänsch
(Berlin),
Z. Chen (Beijing),
G. Dziuk (Freiburg), D. Kessler, R.H. Nochetto (UMD College Park), and
K.G. Siebert (Augsburg).
TetGen, a 3D
tetrahedral mesh generator
based on Delaunay method
TetGen is a C++
program for three-dimensional tetrahedral
mesh
generation. It generates boundary constrained meshes and quality meshes
for 3D piecewise linear domains. It uses Delaunay
based mesh generation
algorithms. It's a public domain program and runs almost all platforms
with a C++ compiler.
This talk discusses the mesh problems TetGen
solves
and the algorithms
TetGen used. Some main features of TetGen are introduced by simple
examples.
Symmetric
Differential Operators on
Polyhedral Manifolds
Whereas
the
traces of differential geometric operators in the discrete
setting (e.g. Laplacian and mean
curvature) are well
understood, the
underlying operators themselves are not as thoroughly studied.
We present a novel simple approach towards the discrete Hessian
and
Weingarten operators by considering a weak (integral) version of
vector-valued normal curvatures on piecewise linear models.
In contrast to the smooth setting where these operators are a priori
symmetric this is not the case in the discrete world. Hence we briefly
discuss their symmetrization using a
certain
1-parameter family of
convolution operators and show that the traces as well as the principal
directions remain unchanged throughout this symmetrizing
family.
In particular, the traces (Laplacian and
mean
curvature) agree
with what has been well established for polyhedral surfaces. We will
shortly discuss how to recover the symmetrization at vertices from the
symmetrization at incident edges.
Adaptive Multilevel
Methods for Optimal
Control Problems
A
function space oriented approach to solving optimal control problems
with interior point methods is presented. The algorithm combines
advantages from both direct and indirect methods by obtaining accurate
solutions without analytical preparation.
Convergence theory and the algorithmic realization via inexact Newton
methods and adaptive refinement are sketched, and ODE examples are
given.
Finally, the prospective extension to PDEs
with
application to
hyperthermia treatment planning is discussed.
Last updated on 23/05/2003 09:21