Here is the Logo.

Parallel Image processing with Mathematica and algorithms in CWeb.

Pages for the documentation of my research at University Leipzig.
Maintained by Patrick Scheibe.
home
projects
interests
cweb
mathematica
people
work
jens-peer
publications
contact

Results of my research

Publications

Poster

Presentations

When Mathematica uses the graphics card - How to use CUDA with Mathematica

This was a presentation I gave at the 2nd Mathematica-day in Leipzig at september 24th. I showed the whole process of integrating the gpu-calculations into Mathematica and used julia-sets and a rigid registration as example. The presentation notebook and the required Mathematica-package can be downloaded here: CUDALecture.tar.gz (5MB)

Diploma Thesis

Abstract

One important application of image processing in medicine is to register tissue samples onto another. Registering these high textured images with non-parametric methods leads sometimes to solutions which are known to be suboptimal.

This thesis is concerned with a novel approach for image registration. We present a projection technique for a curvature based non-parametric registration method which suppresses unwanted vortices in the displacement field. This new strategy does not change the registration method itself but it continuously leads the process of registration to a vortex-free solution.

The grounding method was introduced in [Ami94], used in [BK05] and extended in [BK06] with a vortex suppression term. Our new method calculates a Helmoltz decomposition on the intermediate steps and projects out unwanted vortices. This thesis describes the whole process of the image registration. Starting from the mathematical description of the Helmholtz decomposition, its variational presentation and the consequential partial differential equation, we will go on by looking at the discrete approximation, parts of the implementation and the application on images. Finally the results in comparison with the two other methods of Kuska and Braumann [BK05, BK06] are presented. For this purpose, samples are given and deformed with an artificial transformation. We will use these results and discover general properties, advantages and drawbacks of the different approaches.