Scientific Visualization


Visualization is a key technology to foster exploratory data analysis as it enables the layering of far more data than any other analysis process could. As the visualization is the interface or view into the data, it accommodates better and faster understanding of complex processes and products. Especially in virtual and collaborative workspaces, visualization techniques must be capable of dealing with real-time constraints, both with respect to data analysis, information fusion and rendering aspects.

Research activities in Scientific Visualization cover a variety of different topics common to many areas of science and engineering. Particular interests include the exploration of large-scale data sets as they arise in medical imaging or numerical simulation techniques. In this context our dominant goal is to define Scientific Visualization as a multi-stage pipeline, including data generation and processing techniques as well as mapping and display techniques. Among others, hierarchical approaches and approaches that efficiently exploit parallelism and/or dedicated graphics hardware to interactively generate, process and visualize large-scale data sets are investigated.

In particular, our activities include the following visualization areas:

Volume Visualization

Our goal is to provide interactive, yet high quality volume rendering techniques for large data sets. In particular, we are developing cost effective parallel rendering algorithms on GPUs, and we discovering new algorithms to render unstructured grids and time-varying sequences.

Tensor Visualization

Interactive visualization of diffusion tensor fields is one the most challenging applications in medical imaging. We have developed a particle engine to explore large-scale tensor fields including a number of different visualization options. For about half a million particles, reconstruction of diffusion directions from the tensor field, time integration and rendering can be done at interactive rates. Different visualization options like oriented particles of diffusion-dependent shape, stream lines or stream tubes facilitate the use of particle tracing for diffusion tensor visualization. The proposed methods provide efficient and intuitive means to show the dynamics in diffusion tensor fields, and they accommodate the exploration of the diffusion properties of biological tissue.

Flow Visualization

Besides interactive visualization options for large flow fields, feature extraction and uncertainty visualization in such fields play an important role. Particle tracing including numerically accurate integration schemes has turned out to be a valuable method in this area. The extension to triangular and tetrahedral grids, and the visualization of unsteady flow fields is actually under investigation.

Iso-Surface Extraction and Rendering

By using the latest features of current consumer-level PC graphics cards, we develop novel approaches to indirect volume visualization techniques. Iso-Surface extraction from tetrahedral grids that allows for the interpolation of arbitrary per-vertex values and point-based rendering of high-resolution surfaces from large volumetric data sets and time-varying sequences have been successfully demonstrated.

Terrain rendering

In recent years, the rendering of high resolution terrain data including photo textures has gained increasing attention. In this work we combine the advantages of continuous LOD semi-regular meshes with the advantages of a discrete LOD hierarchy, thus avoiding any re-triangulation of such fields at run-time. The proposed method generates high quality renderings by supporting a continuous LOD representation including photo-texturing. In contrast to previous methods, the terrain is guaranteed to be refined within a user-defined screen- and world-space error. Aliasing is avoided by employing optimal geometry filtering at the best possible geometric resolution. At run-time, discrete sets of decimated mesh structures are transmitted progressively to the GPU, resulting in high bandwidth efficiency.