Efficient Parallel Splatting Algorithms with Sparse Data Structure

• Main Authors: Chih-Hsien Hunag, 黃智顯
• Other Authors: JEN-SHIUH LIU
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/19927903914350018883

碩士 === 逢甲大學 === 資訊工程學系 === 89 === Splatting is a very popular feed-forward direct volume rendering algorithms that help engineers or scientists to visualize scalar and vector fields defined on 3D grids to product final image of high quality.However, the original splatting algorithm must traverse all the voxels in some order. Thereby the high computation cost of original splatting algorithm increases exponentially as the size of the volumetric data sets increases like many other volume rendering algorithm. This thesis reports two new sparse-matrix data structure implemented in original splatting algorithm that enhances the speed of splatting without trading off image quality. These new methods reduce the rendering time and memory storages by employing sparse-matrix data structure that allows to splat only the voxels of interest. It is also reported that the two sparse-matrix splatting algorithm are suitable if the input volume datasets is sparse not dense. The proposed methods were also implemented on an SP2 parallel machine and FCU Cluster system along with two sparse-matrix splatting algorithms and the original splatting algorithm. Finally, the experimental results on several test datasets using let us know that the two sparse-matrix splatting algorithms have better performance among the original splatting algorithm in parallel process