Parallel Volume Rendering Based on LOD Method

Abstract

Parallel volume rendering is an effective approach, which can achieve fast rendering of large-scale data sets by dividing the volume and then assigning them to different cluster nodes for rendering in parallel. This paper presents some acceleration techniques in each stage of the parallel-rendering- pipeline. In the stage of data management, we employ a LOD method to overcome the constraints of storage capacity and to speed up the rendering while maintaining image quality. In the stage of rendering, we use the geometric template method to reduce the huge computational cost of the CPU. Finally the binary-swap algorithm is used in the image synthesis stage. The load-balancing problem is a key issue in parallel rendering and it would become more serious especially when we zoom in or out. After the use of the LOD method, we resolve the load-balancing problem by introducing the kd-tree data structure and utilizing the dynamic rendering time of the last frame as the balancing standard. Several experiments on the Virtual Human (VH) data sets show that these techniques can effectively improve the performance of parallel volume rendering.