EXPLORING GPU- AND CLUSTER-BASED IMPROVEMENTS FOR OVER-SAMPLED VOLUME RAY CASTING OPACITY CORRECTION

Abstract

Performance improvements to the known opacity correction mechanisms for over-sampled volume ray casting (VRC), especially using two forms of commodity hardware, are explored. Data-parallel strategies that enable exploitation of parallelism using either: (1) a programmable graphics processing unit (GPU) or (2) cluster computation are a prime focus. The GPU-based approach is finely granular. The cluster-based approaches here utilize less finely granular processing that allows acceleration through multi-processing and multi-threading. These approaches also include features, such as early ray termination, empty-space skipping, term rearrangement, and term reduction, that have not been previously explored in depth for opacity-corrected VRC. A new strategy enabling more accurate opacity correction is also presented. The performance of the improvements on real volume data are also explored. The improvements allow opacity correction to be performed in a way that efficiently exploits either GPU- or cluster-based capabilities.