GPU ray casting

Abstract

For many applications, such as walk-throughs or terrain visualization, drawing geometric primitives is the most efficient and effective way to represent the data. In contrast, other applications require the visualization of data that is inherently volumetric. For example, in biomedical imaging, it might be necessary to visualize 3D datasets obtained from CT or MRI scanners as a meaningful 2D image, in a process called volume rendering. As a result of the popularity and usefulness of volume data, a broad class of volume rendering techniques has emerged. Ray casting is one of these techniques. It allows for high quality volume rendering, but is a computationally expensive technique which, with current technology, lacks interactivity when visualizing large datasets, if processed on the CPU. The advent of efficient GPUs, available on almost every modern workstations, combined with their high degree of programmability opens up a wide field of new applications for the graphics cards. Ray casting is among these applications, exhibiting an intrinsic parallelism, in the form of completely independent light rays, which allows to take advantage of the massively parallel architecture of the GPU. This paper describes the implementation and analysis of a set of shaders which allow interactive volume rendering on the GPU by resorting to ray casting techniques.