discrete samples on a 3D grid -- by integrating illumination along rays through the volume. The optical
properties in the volume are derived from volume data using so called transfer functions. In addition to
simple emission-absorption models, scattering can also be taken into account but drastically increases
computational cost.
The talk consists of two parts: First I present a fast GPU implementation of direct volume rendering,
that exploits the functionality of modern graphics processors but has to traverse the entire volume. The
second part suggests a hierarchical linear approximation of the volume data that allows for computing
strict error bounds of the integrals along the rays through volume. These error bounds allow us to
traverse the volume at much coarser levels while guaranteeing high quality results. This approach will
greatly reduce computational cost as well as memory bandwidth. This will be of great importance in
the near future given the quickly rising resolution of volume data sets.