On the (Un)Suitability of Strict Feature Definitions for Uncertain Data - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 42

Chapter 4
On the (Un)Suitability of Strict Feature
Definitions for Uncertain Data

Tino Weinkauf

Abstract We discuss strategies to successfully work with strict feature definitions
such as topology in the presence of noisy/uncertain data. To that end, we review
previous work from the literature and identify three strategies: the development of
fuzzy analogs to strict feature definitions, the aggregation of features, and the filtering
of features. Regarding the latter, we will present a detailed discussion of filtering
ridges/valleys and topological structures.

4.1 Introduction

Features are not only ubiquitous in scientific data, but they are inherent to the
underlying natural phenomena. For example in fluid dynamics, vortex structures
influence important properties such as the lift of an airfoil or the drag of a car. Under-
standing such features—when and where they occur, their strength, their dynamics—
is crucial to understanding and controlling the underlying phenomena.
Definition and interpretation of features depend on the underlying application, but
usually they represent important structures (vortex, stagnation point) or changes to
such structures (events, bifurcations). There are at least two ways of defining features
in scientific visualization:
￿

Smooth Feature: a fuzzy area of the domain where every point adheres to a given
definition to some extent .

￿ Strict Feature: a well-defined subset of the domain which fully adheres to a given

definition; usually a geometric object such as a point, line, or surface.
Smooth feature definitions lend themselves to interactive visual analysis approaches
such as SimVis [ 1 ], where the user can create a fuzzy feature definition interactively
by brushing in different views of the data and exploring the result in linked views.
The result is subjective, but the approach leaves room for exploration in cases where
the features cannot be described a priori (yet). Smooth feature definitions address
uncertainty or noise, at minimum, by communicating it through their fuzziness.

( B )
Max-Planck-Institut für Informatik, Saarbrücken, Germany
e-mail: [e-mail]
© Springer-Verlag London 2014

T. Weinkauf