Appearance Preserving Data Simplification

Abstract

Many visualization environments constantly face the issue of dealingwith large, complex datasets. Often these datasets are so complexthat rendering a visualization would seem impractical. Likewise,enormous amounts of data may overwhelm the human visual system; therebyrendering the data incomprehensible. Thus, the need arises to deal withthese datasets in some arbitrary manner such that the resultingdataset represents the original whole --- while reducing thecost on the human and computer visual system.

A closely related problem can be found in geometric models, typicallyrepresented as a piecewise linear collection of connected polygons (amesh). Meshes can be obtained from range scanners or created with acomputer aided design package. However, these obtained meshes areoften very dense and have high spatial frequency. An active area ofcomputer graphics research is directed at the simplification of thesedense meshes. Initially, mesh simplification research aimed atpreserving only the topology, but the most recent research, appearancepreserving mesh simplification, is aimed at simplification whilepreserving surface properties of the mesh, such as color or texture.

Our work addresses the use of appearance preserving meshsimplification in a data simplification environment, as well as, theissues of doing so. As a result, we present and demonstrate a generalmethod to simplify large multidimensional datasets using anyappearance preserving mesh simplification algorithm. We add the use ofprincipal components analysis to reduce the dimensionality of the dataprior to simplification, which allows faster simplification on highdimensional data, and despite the reduction in dimensionality we haveshown full preservation of key features in the dataset. In addition,we introduce spatial locks to preserve important data elements duringthe simplification process.