Abstract

Multiresolution techniques for volume visualization provide means for interactive exploration of very large data sets. This work provides a multiresolution texture-based technique that uses an adaptive scheme to render a portion of a given volumetric data set in a region of interest at a high resolution. Those portions of the volume away from the region of interest are rendered at progressively lower resolutions.

The algorithm is based on the segmentation of texture space into an octree of texture elements (texels), where the leaves of the tree define the original data, and the internal tiles define lower-resolution versions. Rendering is done adaptively by selecting high-resolution tiles close to a center of attention and low-resolution tiles away from this area.

To understand the essence of this technique, consider the image to the left. The original grid consists of 256 tiles. A tile is selected when the distance from the center of the tile to p is greater that the diagonal size of the tile.

The "selected" tiles for each resolution are shown in gray. Here, the original grid contains 256 texels, the final image, if done in a multiresolution way, could contain 49 texels -- one from level three, seven from level two, 13 from level one, and 19 from level zero.

(The "viewpoint" p is indicated in the final image.)

We limit artifacts introduced by this method by using viewpoint-centered spherical shells. These shells are shown on the left. The shells intersect the solid texture and the system maps the texture onto the shells.

We modify the transfer/opacity function in lower resolution areas so that we minimize differences between optical properties between low- and high-resolution areas. This produces a smooth image. With this technique, we can view a volume of data from an arbitrary viewpoint and produce interactive fly-throughs of the data. The solid textures representing the data are transfered to the imaging system in a multiresolution way and blended using the modified transfer/opacity functions.

The operation of the algorithm is shown for a trebecular bone data set. The data set consists of 256x256x256 voxels. The upper image shows the full data set, and the lower shows a multiresolution version.

It is possible to use this technique to produce viewpoint-dependent renderings of very large data sets.

Publications

• Eric C. LaMar, Mark A. Duchaineau, Bernd Hamann, Kenneth I. Joy, Multiresolution Techniques for Interactive Texture-Based Volume Visualization, in: Visual Data Exploration and Analysis VII, pp. 365--374, 2000.

Contact