| Volume visualization is computationally expensive. While software solutions are far too slow for real-time rendering of the data sets, costly hardware only effectively handles small data sets. Usually, excessively expensive large-scale parallel systems are needed for interactive volume visualization applications. Interactive volume visualization on a remote desktop is unsolved and challenging. My dissertation presents a framework and a working prototype system providing highly interactive (30 frames/second) remote visualization of volume data sets, without duplicating the data sets on the client side.; Our framework involves a back-end rendering server, a fast network link and a desktop front-end client. First, we have developed an optimized point primitive, FastSplats. Using FastSplats, our software image-aligned sheet-based (IASB) splatting renderer avoids the systematic performance bottleneck that hardware based splatting renderers inevitably encounter when high-end features, such as per-pixel classification/shading and occlusion culling based accelerations are sought after. FastSplats also support parallelizations of IASB splatting algorithms on a variety of parallel computing platforms. Second, as one of the first few efforts in parallelizing IASB splatting, we found out that for different types of data sets, with or without heavy occlusion, there are different ways to render them with high scalability. Specifically, we devised a parallel rendering scheme for data sets of heavy occlusion. Third, with an efficient parallel back-end rendering server implemented, we use an image-based representation to support user interaction on the client side. The back-end rendering server renders each view of a data set into an image-based representation, which is sent to the front-end client. The client viewer uses the image-based representation to mimic user interaction for nearby views. This framework is novel and significant, in that it decouples the expensive volume rendering from user navigation, allowing the user to stay focused on comprehending the data set without disruptions in interaction due to latencies. The re-use of the image-based representations allows for an order of magnitude reduction in the network bandwidth as compared to streaming videos of visualization.; Hence, a 30 frames/second rate on the client side does not require a 30 frames/second with back-end rendering or network communication. This is a framework providing a progressive refinement. As the user is navigating in the data sets on the client side, image-based rendering provides a smooth sense of interaction to the user. As soon as the user stops movement and focuses on a static view to inspect for details, the correct view gets rendered on the back-end server and sent to the client side and shown on the screen in seconds. |
