Splatting II: High-Quality Volume Rendering Without The Wait

                           Dr. Klaus Mueller 
                        The Ohio State University

Abstract:

The majority of today's 3D graphics systems represent objects as meshes of
polygons, which can be rendered at high frame rates. The polygonal
representation performs well in many settings, however, certain real-world
phenomena and interactions exist that are not easily modeled with surface-based
methods. For example, semi-transparencies and subtle micro-structures, such as
 soft tissue in medical datasets, as well as natural phenomena, such as clouds
and fog, are hard to represent with a reasonable number of polygons.
Interactive operations, such as cutting and sculpting, are also difficult with
surface rendering. Volume rendering, on the other hand, retains the 3D raster
representation of the volumetric object, obtained via MRI, CT, scientific
simulations, procedural models, or cryosectioning, and allows these operations
and effects to be realized in an intuitive way.
Volume rendering is considerably more complex than surface rendering, and this
impedes speedy image generation. One of the more efficient volume rendering
algorithms is the Splatting technique. However, while Splatting provides images
of reasonable quality for still frames, a variety of artifacts appear in
animated and interactive viewing. We address all of these issues in the new
 generation of Splatting: Splatting II. This new variant not only generates
images of high quality, but also renders them fast, due to an efficient data
culling process.
We then give a sneak preview of Splatting II+, a parallel implementation of
Splatting II that decouples the parallel rendering process from the process
that combines the partial parallel rendering results for display. This gives
way to an interactive volume visualization system that is tolerant to network
and rendering latencies, and combines partial rendering results as needed for
the current viewpoint.
Finally, we point out that insight gained in volume rendering can also be
utilized for computed tomography (CT). CT can be considered an inverse of
volume rendering: It reconstructs a volumetric object from its projections. We
will discuss a popular CT method, the Algebraic Reconstruction Technique (ART),
and improvements we have achieved for its use in cone-beam reconstruction.