ERCIM News No.25 - April 1996 - CNR
TAn ­p; A New Multiresolution Volume Visualization System
by Claudio Montani, Enrico Puppo and Roberto Scopigno
The main problems in volume dataset visualization are the excessive data
complexity, which limits interactivity, and the lack of generality and integration
of the rendering solutions. Moreover, Volume Visualization systems often
manage only a single class of data (e.g. regular or curvilinear). TAn (Tetrahedra
Analyzer) is a new prototypal system for Volume Visualization, designed
to cope with data complexity. TAn supports multiresolution modelling and
visualization of volume datasets by adopting an innovative representation
based on tetrahedral decomposition. No limitations are imposed on the classes
of data that can be managed: TAn processes regular, curvilinear and scattered
data. The tetrahedral decomposition adopted is also an effective choice
for the design of integrated visualization methods.
Volume Visualization is now recognized as a major subfield of Scientific
Visualization. This discipline entails the management of datasets generically
defined as the result of a measure over a discrete set of sites in the 3D
space. Volume datasets are usually classified according to the distribution
of the sampling sites in the 3D space. A number of techniques have been
proposed to visualize such data efficiently.
A problem common to all classes of datasets is the huge amount of data involved;
this affects both storage requirements and visualization times. Dataset
complexity has been taken into account in a number of recent proposals to
reduce visualization times, but the techniques developed are designed mainly
for regular volume datasets.
A more general approach consists in using approximated representations of
the dataset, defined on meshes at reduced resolutions, in order to allow
fast visualization when high accuracy is not needed. The idea is to work
on data simplification rather than on graphics output simplification. Given
a rule for simplifying the dataset, a multiresolution representation can
be simply achieved by applying this rule with different approximation parameters;
this can effectively improve the efficiency of data rendering.
Naive ways to achieve multiresolution, like subsampling or averaging have
several limitations: they only control the size of the reduced dataset but
not its accuracy with respect to the original dataset; they are not adaptive,
i.e., they do not permit a variable data density to be maintained over different
regions of the domain, according to the variations of the scalar field represented;
they are only suitable for regularly gridded data.
In a recent paper, we proposed a methodology for building and manipulating
volume representations at multiple resolution. Our multiresolutions model
is based on tetrahedral decompositions with scattered vertices, which can
be obtained from any initial dataset. Tetrahedral meshes have been widely
used to represent irregular volume data. In the context of volume visualization,
a number of rendering techniques have been proposed which are well-suited
for tetrahedralized datasets.
Multiresolution is achieved through a compact model that incorporates a
whole sequence of tetrahedral decompositions, corresponding to increasingly
accurate approximations of the original dataset. Given a threshold
for the approximation error, the model is able to provide a tetrahedralization
based on a minimally sufficient subset of the sites that achieves an accuracy
of in approximating the original dataset. The model is built through an
adaptive incremental approach driven by local coherence. The key idea behind
our model is that a high number of different decompositions at increasing
accuracy can be obtained on the basis of a moderate number of tetrahedral
cells. Different tetra-hedralizations are not stored independently: a unique
data structure embeds them all, and a simple but efficient algorithm extracts
each single model on-line at any arbitrary tolerance .
A prototypal volume visualization system, TAn (Tetrahedra Analyzer), was
designed which adopts the proposed multiresolution tetrahedral representation:
a compact representation scheme which stores all the intermediate steps
of a progressively refining 3D triangulation process. Dataset triangulation
is performed only once, as a preprocessing phase. The results of this incremental
refinement process are saved on a file (known as the history file). At run
time, TAn allows the user to choose the tolerance; a tessellated representation
of the dataset which satisfies the specified tolerance is then extracted
from the history data.
Projective rendering approaches have been adopted.
The main features of TAn are:
- construction of the multiresolution representation based on tetrahedral
- selection of the levels of detail (lod) used in rendering
- loading and interactive editing of the transfer function (i.e. the
function which converts field values into colour and opacity)
- data visualization, supporting multiple rendering approaches: isosurface
fitting and rendering; direct volume rendering (DVR), adopting the Projected
Tetrahedra method; approximated hybrid rendering (isosurfaces and DVR)
- multiresolution visualization: a progressive rendering modality is
available (the low resolution lod is rendered in the highly interactive
phases, the high resolution lod otherwise).
The TAn system works on SGI workstations and adopts OpenGL and the XForms
user interface toolkit for graphics output and GUI design.
TAn is the result of a scientific cooperation between three CNR Institutes:
CNUCE and IEI, Pisa, and IMA, Genoa.
The TAn system has been released in the public domain. The software, compiled
for SGI workstations, and a downloadable preprint of the paper that describes
the data representation scheme adopted are available on our World Wide Web
site at URL:
Our research on volume data management and rendering will continue. We are
working on the extension of some of the techniques provided in the current
- testing of an enhanced triangulator which provides correct management
of non-convex datasets through the evaluation of both data and warping errors
- improvement of the hybrid (isosurfaces and DVR) rendering method
- design of effective multiresolution visualization techniques, which
can provide different levels of detail over different portions of the domain
in a single image (for example, to allow focus area selection and visualization).
Roberto Scopigno ­p; CNUCE-CNR
Tel: +39 50 593304
return to the contents page