Applications of data compression to three-dimensional scalar field visualization
Paul Ning · 1993
Scientific visualization is the use of computer graphics techniques for analyzing and displaying experimental or simulated data, and has been promoted as an efficient tool for understanding large, multi-dimensional datasets that arise in many applications. The use of visualization does not eliminate, however, the problem of data management; rather, data management issues must be addressed by the visualization tools themselves. Thus, it is only natural to apply data compression techniques in order to improve the performance of visualization systems. In this dissertation, we demonstrate how compression can be beneficial to both isosurface generation and volume rendering, two common techniques for displaying volumetric (3-D) scalar fields. Isosurface generation involves the construction of a polygonal model which approximates a contour surface in the volume. The number of polygons generated can be quite large, however, which increases rendering time and storage costs. We develop a variable-resolution isosurface generation algorithm which constructs accurate surface models over a wide range of compressed polygon counts. This algorithm imposes an octree hierarchy on the volume data, and applies tree pruning techniques to find good surface models at lower resolutions. Instead of extracting and displaying a single surface, volume rendering treats the entire 3-D scalar field as a collection of sources and attenuators, and integrates these contributions along the viewing direction to form a projected image of the translucent volume. A volume renderer requires access to all of the scalar field samples, as well as to several derived quantities. For large datasets, the storage used by these methods can be very high, and rendering speed is generally slow. We introduce a compressed volume format based on vector quantization that provides storage savings, and also may be exploited for faster rendering.