| This dissertation contains our contributions in methods to speed up volume rendering, an important subfield of scientific visualization. We develop a framework composed of a system and a set of algorithms for handling large datasets of various forms (e.g., regular and irregular volumetric grids). A special emphasis of our framework is on the development of practical parallel algorithms for visualization.; For the regular grid case, where research of efficient techniques is fairly advanced, we propose a parallelization of a known rendering algorithm. Our major contributions in this case are the introduction of content-based load balancing and the pipelined compositing approach. We present the new algorithms and their implementation.; For irregular grids, we propose a fast rendering algorithm. Our method uses a sweep-plane approach to accelerate ray casting, and it can handle disconnected and nonconvex (even with holes) unstructured irregular grids with a rendering cost that decreases as the disconnectedness decreases. The algorithm is carefully tailored to exploit spatial coherence even if the image resolution differs substantially from the object space resolution. We establish the practicality of our method through experimental results, and we also provide theoretical results, both lower and upper bounds, on the complexity of ray casting of irregular grids. Our work in irregular grids also includes a proposal for a parallelization of the method for distributed-memory machines.; Our final contribution is in the simplification of irregular grids. Because the size of the grids can sometimes be overwhelming, we discuss the simplification problem to approximate representations of irregular grids. We present a complete solution for the two-dimensional case (e.g., height-field terrains), and preliminary work on extensions to general polyhedral surfaces and three-dimensional irregular grids. |
