Volume Visualization and Visual Queries for Large High-Dimensional Datasets

Abstract

We propose a flexible approach for the visualization of large, high-dimensional datasets. The raw, highdimensional data is mapped into an abstract 3D distance space using the FastMap algorithm, which helps, together with other linear preprocessing steps, to make changes to the resulting 3D representation within a few seconds. Thus exploration of such datasets is a less tedious task compared to other techniques. We use volumes with four components to enable the user to brush an attribute selection onto the volume for inspection. We exploit multiple transfer functions for displaying these attributes and also to filter one attribute with values of another. An advantage of this volume sampling approach is that the rendering performance is independent of the dataset size. The drawback of limited resolution can be overcome by providing a linked detail view for a freely selectable portion of space. Examples of the inspection and filtering possibilities using a silvicultural dataset illustrate the strengths of our approach.