Efficient methods for volumetric graphics
Arie E. Kaufman, Roni Yagel · 1991
Driven by the prospect of three dimensional rasters as a primary vehicle for future 3D graphics and volumetric imaging, we aim at devising software and hardware tools needed to make this foresight a reality. The major contributions of this work are the design of the Flipping Cube Architecture, the development of the template-based approach to volume viewing, and the combination of both into a special-purpose system for real-time high-resolution volume visualization. In the Flipping Cube Architecture, N$\sp3$ volume elements (voxels) are distributed among N slice-oriented memory modules in a scheme that allows real-time viewing from one third of the possible viewing directions. The aggregate of voxels comprising a viewing ray is fetched in parallel, one voxel from each memory module, and fed into a special purpose mechanism for ray projection. In order to achieve real-time viewing from arbitrary direction the system employs a shift mechanism that accomplishes real-time re-orientation (flipping) of the data inside the modular memory by utilizing either a special purpose Flip Buffer or a 3D shearing method. These two flipping methods are compared in terms of performance, memory requirement, and hardware realization. We analyze the expected performance of our system and conclude that it has the potential to support, for the first time, real-time high-resolution volume viewing from an arbitrary direction. We present a template-based volume viewing algorithm that exploits ray-coherency in parallel viewing in order to expedite rendering. The algorithm generates a single ray and uses it as a template for all other rays, saving the need for repeated ray generation. We demonstrate its superior performance and describe how it can be extended to support screen supersampling, continuous rays, and adaptive traversal. Our three-phased approach guarantees uniform space sampling with discrete rays by carefully placing the ray origin and by employing discrete space arithmetic for ray-volume intersection calculation. Next, we show how the template-based parallel viewing algorithm finds a surprisingly efficient realization as part of the Flipping Cube Architecture while maintaining real-time throughput.