High Quality Template-Based Volume Rendering

Volumetric representation of 3D scenes is becoming widely used in scientific visualization and computer graphics. Volume rendering confronts performance hurdles due to the voluminous nature of the data, and quality hurdles due to its discreteness. In this paper we present a three-phase template-based algorithm for expediting parallel viewing of volumetric datasets. Our approach exploits coherency between rays in parallel views by storing, in a ray-template, information that is common to all rays. From an axis-aligned plane, and within the projected region of the volume, parallel rays are cast into the volume by repeating the sequence of steps specified by the ray-template. We show that the template can also store information that is used to expedite screen-space and object-space supersampling. Finally, we utilize templates to support adaptive volume traversal. 1: Introduction Visualization is the process of converting complex information to a format that is amenable to human understanding, while maintaining the integrity and accuracy of the information. One type of information that can benefit from visualization techniques is volumetric data which consist of information on three dimensional phenomenon. Volumes are commonly composed of a set of values sampled or computed at the integral points of a 3D Euclidean grid. The process of determining, for a given view point, which parts of the volume are visible is called volume viewing. Algorithms for volume viewing operate by scanning through the data points, called voxels, in a strict order designed to achieve hidden voxel removal. Volume viewing algorithms can be classified according to the partial voxel occupancy they support. Many early algorithms [Herman and Liu 1979, Reynolds, Gordon, and Chen 1987, Tuy and Tuy 1984] assume uniform (binary) occupancy, that is, a voxel is either fully occupied by some object or it is devoid of any object presence. In contrast to uniform occupancy, methods based on partial occupancy assume that data values represent partial voxel occupancy by an object of homogeneous material. This provides a mechanism for the representation and display of objects that are smaller than the acquisition grid or that are not aligned with