06/01/1998 09:00 AM
Computer Science
There has been significant work done in the field of surface simplification to reduce the amount of geometry that the renderer has to work on. Very little has been done to reduce the amount of geometry in a volumetric mesh so that faster volume rendering rates can be achieved. In this work we have proposed a model that allows decimation of volumetric meshes. Our model treats a volumetric grid as a volume of some material whose density values are defined at the vertices. Our method generates successive approximations of the original volume based on local changes in density. We use a direct volume rendering system based on a generalized software scan conversion of polygons, rather than the more conventional ray-casting, projection, or splatting of cells. A polygon-based volume renderer does not concern itself with the geometry of cells and, strictly speaking, cares only about polygons, which don\'t have to be connected to form cells. Hence the only geometry involved in rendering is polygons and the problem of decimating a volume is reduced to choosing a subset of the polygons in the volume. We examine these polygons and throw away those that are not significant according to our error metric. A great advantage of this approach is that we do not have to regrid the data. The decimation method without regridding proves useless for volumes decimated more than 50% because it creates holes in the volume which result in objectionable artifacts in the volume rendered images. We have also implemented a decimation method that maintains a valid tetrahedralization at all times during decimation, and allows us to decimate more than 70% of the volume without much compromise in image quality. The advantage of regridding is that the decimated volume can be used for other applications like isosurface generation and streamline computation.
