We have covered in the previous sections several algorithms that achieve ``fast'' volume rendering. Most of these are non-trivial parallel algorithms that try to run as efficient as possible on a given architecture. Figure 11 summarizes the algorithms described in this report.
Figure 11: Summary of algorithms described in the previous sections.
Most of the algorithms presented in this paper belong to the same
category of performance, they can generate very fast pictures but are
not suited for real-time volume rendering. One can use these
algorithms to generate sequences of animation but not ``explore'' data
using them. The only algorithm presented that was developed for
real-time user interaction was the one by Schröder and
Salem
, but the requirements for the
algorithm are very high in computing power and the image quality
is low.
Our goal is therefore to continue research on this area, trying to develop techniques that make exploration of volumetric data possible. We are currently involved in the CUBE project that, when fully implemented, will be able to achieve performance rates over two order of magnitudes faster than the CM-2 algorithm just presented for a hundredth of the cost.
Other direction of research is to developed and implement algorithms for parallel machines. We are currently developing a small prototype based on previous code developed by the members of the CUBE research group for profiling purposes. The small term goal is to get fast, non-interactive performance on the Intel iPSC/860. When the new Intel Paragon comes in early next year we hope to be able to achieve performance measures better than the CM-2 implementation.