External Memory Techniques For Isosurface Extractions In Scientific Visualization

Isosurface extraction is one of the most effective and powerful techniques for the investigation of volume datasets in scientific visualization.
Previous isosurface techniques are all main-memory algorithms, often not applicable to large scientific visualization applications. In this
paper we survey our recent work that gives the first external memory techniques for isosurface extraction. The first technique, I/O-filter, uses
the existing I/O-optimal interval tree as the indexing data structure (where the corner structure is not implemented), together with the isosurface
engine of Vtk (one of the currently best visualization packages). The second technique improves the first version of I/O-filter by replacing the
I/O interval tree with a metablock tree (whose corner structure is not implemented). The third method further improves the first two, by using
a two-level indexing scheme together with a new meta-cell technique and a new I/O-optimal indexing data structure (the binary-blocked I/O interval tree)
that is simpler and more space-efficient in practice (whose corner structure is not implemented). The experiments show that the first two methods perform
isosurface queries faster than Vtk by a factor of two orders of magnitude for datasets larger than main memory. The third method further reduces the disk
space requirement for 7.2-7.7 times the original dataset size to 1.1-1.5 times, at the cost of slightly increasing the query time; this method also exhibits a
smooth trade-off between disk space and query time.