Volume Rendering: Intro / Pre-Integration / Bonsai / Extensions
Nowadays the enourmous leap in 3D graphix hardware allows us to facilitate volume rendering even on low end computer systems. In the past most hardware accelerated algorithms have focused on the utilization of 3D-textures as supported on a variety of powerful but expensive SGI workstations. Another approach is to use cell-projecting algorithms, which were introduced by Shirley and Tuchman in 1988. These can operate on both structured and unstructured grids and are well suited for implementation on cheap graphix hardware. In comparison to existing volume renderers cell-projection algorithms did not support fast color lookup tables, which are essential for medical volume visualization. By excessively exploiting the texture mapping features of todays 3D graphix accelerators, we were able to show that the application of transfer functions is possible in every rendering framework based on cell-projection. Additionally, we were able to extract multiple shaded isosurfaces without explicitely reconstructing the iso surfaces geometry. This is especially interesting, because the rendering times do not depend on the number of isosurfaces. In this context the problem of mixing an isosurface patch with a volume cell has been solved as well (see Vis 2000 paper).
Transfer Functions and Iso Surfaces (Gallery 1):
The left image shows a synthetic test data set with a partially transparent transfer function, which maps a portion of the density values of the volume from blue (low) to green (high) resulting in the transparent ring in the center. In the right picture a linear transfer function was used to render the volume. It has been mixed with 10 smoothly shaded isosurfaces of different colors, only 5 of which are visible.
Pre-Integration (Gallery 2):
The rightmost two images are examples of a pre-integrated isosurface extraction and blending texture. Three isosurfaces can be extracted simultaneously with these textures, but as long as the texture resolution is high enough there is no upper limit on the maximum possible number of isosurfaces. The next two pictures are a 3D view of the ray integral given in our paper and its two-dimensional approximation. This precomputed integral was used to render the bluntfin data set that is depicted below in the middle image.
Volume Data Examples (Gallery 3):
Now to some real world data sets - from left to right: an MRI head scan, the curvilinear NASA bluntfin data set, and a CT scan of a bonsai with kind permission of Dr. Bernd Tomandl from the Division of Neuroradiology of the Department of Neurosurgery at the University of Erlangen. The volumetric data sets of the bonsai (and two other bonsais) are freely available for scientific purposes. When using the data sets please cite this URL as a reference. The bonsai and a variety of other well known volumetric data sets are also available at the data section of the VolVis home page.
Extensions to the Original Algorithm (Gallery 4):
Since the pre-integration of the ray integral takes quite some time we devised a hardware-accelerated approach that is able to recompute the pre-integration table at interactive speed. As a consequence, one can explore unstructured volume data sets comfortably by manipulating the transfer functions interactively (see publications). In addition, our method has been improved to run on the PC platform (left and middle) and on low-end graphics adapters (right).