Accelerated Splatting using a 3D Adjacency Data Structure

We introduce a new acceleration to the standard splat-ting volume rendering algorithm. Our method achieves full colour (32-bit), depth-sorted and shaded volume rendering significantly faster than standard splatting. The speedup is due to a 3-dimensional adjacency data structure that efficiently skips transparent parts of the data and stores only the voxels that are potentially visible. Our algorithm is robust and flexible, allowing for depth sorting of the data, including correct back-to-front ordering for perspective projections. This makes interactive splatting possible for applications such as medical visualizations that rely on structure and depth information. 1. Introduction The simplification of three-dimensional (3D) volu-metric data sets is a critical step in the exploration of the underlying data. The display of key features of the volume (e.g. iso-regions for scalar data sets [5, 16], streamlines , stream-surfaces and stream-volumes for vector data sets [2]) allows for better comprehension of a 2D projection of the volume. Rendering only those key features allows faster drawing speeds, and also helps prevent important information from being obscured by less important clutter. This enables interaction with the volume, enhancing the 3D understanding of the data. Iso-surface extraction, as introduced by Lorensen et. al. [16] and other researchers [10, 31] is still one of the most popular visualization techniques for the display of volumetric data sets. It displays a 3D object by rendering a surface that represents a constant intensity. This might include one or two transparent or translucent surfaces. However, iso-surface extraction cannot display certain qualities such as light attenuation due to opacity and depth. These can only be accomplished through direct volume rendering algorithms. How do we efficiently skip the unnecessary part of the data while still maintaining the full context of the