Efficient Parallel Splatting Algorithm

Abstract

Splatting is a technique for reconstruction in feed-forward volume rendering, which consumes large portion of time in the entire rendering process. In this work, we exploit object space coherence in the splatting process. Sparse-matrix data structure is introduced to take advantage of object space coherence in order to save computation time and memory storage. With the spatial data structure, no computation is performed for those uninteresting voxels during the splatting process. It is known that volume rendering may take several seconds to hours to render a 3D dataset on a typically single processor. We also present data-parallel volume rendering algorithms in order to achieve real time applications. Experiments are conducted to assess our proposed schemes. Results show that the proposed sparse-matrix algorithm performs very well.