See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221025525

Fluid Flow Visualization

Conference Paper · January 1991

DOI: 10.1007/978-3-642-77165-1_1 · Source: DBLP

CITATIONS READS

72 10,395

2 authors:

Frits H Post Theo van Walsum

Delft University of Technology Erasmus MC
170 PUBLICATIONS   3,934 CITATIONS    187 PUBLICATIONS   4,406 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

BSc Thesis View project

Prosthetic Heart Valve in CTA View project

All content following this page was uploaded by Frits H Post on 10 June 2014.

The user has requested enhancement of the downloaded file.

 FLUID FLOW VISUALIZATION
Frits H. Post, Theo van Walsum

Delft University of Technology, The Netherlands*

Published in: Focus on Scientific Visualization, H. Hagen, H. Müller, G.M. Nielson (eds.),
Springer Verlag, Berlin, 1993, pp. 1-40 (ISBN 3-540-54940-4)

Abstract
This paper presents an overview of techniques for visualization of fluid flow data. As a starting
point, a brief introduction to experimental flow visualization is given. The rest of the paper
concentrates on computer graphics flow visualization. A pipeline model of the flow visualization
process is used as a basis for presentation. Conceptually, this process centres around visualization
mapping, or the translation of physical flow parameters to visual representations. Starting from a
set of standard mappings partly based on equivalents from experimental visualization, a number of
data preparation techniques is described, to prepare the flow data for visualization. Next, a number
of perceptual effects and rendering techniques are described, and some problems in visual
presentation are discussed. The paper ends with some concluding remarks and suggestions for
future development.

1 Introduction
For centuries, fluid flow researchers have been studying fluid flows in various ways, and today
fluid flow is still an important field of research. The areas in which fluid flow plays a role are
numerous. Gaseous flows are studied for the development of cars, aircraft and spacecrafts, and
also for the design of machines such as turbines and combustion engines. Liquid flow research is
necessary for naval applications, such as ship design, and is widely used in civil engineering
projects such as harbour design and coastal protection. In chemistry, knowledge of fluid flow in
reactor tanks is important; in medicine, the flow in blood vessels is studied. Numerous other
examples could be mentioned. In all kinds of fluid flow research, visualization is an key issue.

1 . 1 Purposes and Problems of Flow Visualization

Flow visualization probably exists as long as fluid flow research itself. Until recently,
experimental flow visualization, as described in section 2, has been the main visualization aid in
fluid flow research. Experimental flow visualization techniques are applied for several reasons:
• to get an impression of fluid flow around a scale model of a real object, without any
calculations;
• as a source of inspiration for the development of new and better theories of fluid flow;
• to verify a new theory or model.
Though used extensively, these methods suffer from some problems. A fluid flow is often
affected by the experimental technique, and not all fluid flow phenomena or relevant parameters
can be visualized with experimental techniques. Also, the construction of small scale physical
models, and experimental equipment such as wind tunnels are expensive, and experiments are time
consuming.

* Department of Technical Informatics, Zuidplantsoen 4, 2628 BZ Delft, The Netherlands. Email:

frits.post@cs.tudelft.nl

1
 Recently a new type of visualization has emerged: computer-aidedvisualization. The increase
of computational power has led to an increasing use of computers for numerical simulations. In the
area of fluid dynamics, computers are extensively used to calculate velocity fields and other flow
quantities, using numerical techniques to solve the governing Navier-Stokes equations. This has
led to the emergence of Computational Fluid Dynamics (CFD) as a new field of research and
practice.
To analyse the results of the complex calculations, computer visualization techniques are
necessary. Humans are capable of comprehending much more information when it is presented
visually, rather than numerically. By using the computer not only for calculating the numerical
data, but also for visualizing these data in an understandable way, the benefits of the increasing
computational power are much greater.
The visualization of fluid flow simulation data may have several different purposes. One
purpose is the verification of theoretical models in fundamental research. When a flow
phenomenon is described by a model, this flow model should be compared with the ‘real’ fluid
flow. The accuracy of the model can be verified by calculation and visualization of a flow with the
model, and comparison of the results with experimental results. If the numerical results and the
experimental flow are visualized in the same way, a qualitative verification by visual inspection can
be very effective. Research in numerical methods for solving the flow equations can be supported
by visualizing the solutions found, but also by visualization of intermediate results during the
iterative solution process.
Another purpose of fluid flow visualization is the analysis and evaluation of a design. For the
design of a car, an aircraft, a harbour, or any other object that is functionally related with fluid
flow, calculation and visualization of the fluid flow phenomena can be an powerful tool in design
optimization and evaluation. In this type of applied research, communication of flow analysis
results to others, including non-specialists, is important in the decision making process.
In practice, often both experimental and computer-aided visualization will be applied. Fluid
flow visualization using computer graphics will be inspired by experimental visualization.
Following the development of 3D flow solution techniques, there is especially an urgent need for
visualization of 3D flow patterns. This presents many interesting but still unsolved problems to
computer graphics research. Flow data are different in many respects from the objects and surfaces
traditionally displayed by 3D computer graphics. New techniques are emerging for generating
informative images of flow patterns; also, techniques are being developed to transform the flow
visualization problem to display of traditional graphics primitives.

1 . 2 Overview
The main purpose of this paper is to give an introduction to fluid flow visualization with computer
graphics. It was primarily written for the computer graphics students interested in scientific
visualization, and therefore some knowledge of basic computer graphics concepts and techniques
is assumed. No knowledge of fluid dynamics is assumed; readers interested in the principles of
fluid dynamics theory are referred to text books on fluid dynamics (such as Batchelor, ‘67). This
paper will also be useful for CFD specialists who want to know more about techniques available to
visualize their data. An attempt has been made to provide a connection of the visualization process
with the CFD simulation process. For basic graphics techniques they should refer to text books on
computer graphics (such as Foley et al., ‘90).
Before we turn to computer-aided visualization flow techniques, we will briefly discuss some
concepts and methods from experimental flow visualization (section 2). In section 3, a pipeline
model of the flow visualization process is introduced, and two important stages of the pipeline are

2
described: visualization mapping and data preparation and analysis techniques. Section 4 describes
computer graphics presentation techniques, based on a set of standard forms of flow visualization
described in section 3. Some additional topics in rendering and animation will also be discussed.
Finally, in section 5, some conclusions are drawn and directions for research are indicated.

2 Experimental Flow Visualization

Experimental flow visualization has a long history, and today a wide variety of techniques to
visualize fluid flows is known. Merzkirch (‘87) and Yang (‘89) give an overview of the
fundamental techniques for flow visualization and describe a number of applications of these
techniques. Van Dyke (‘82) shows many excellent pictures of experimental fluid flow
experiments.
Following Merzkirch (‘87), three basic types of visualization can be distinguished: adding
foreign material (2.1), optical techniques (2.2), and adding heat and energy (2.3).

2 . 1 Addition of Foreign Material

Time lines, streak lines and path lines play an important role in experimental fluid flow
visualization techniques:
• Time lines are lines that, once released in the fluid, are moved and transformed by the fluid
flow. The motion and formation of the line, which is often released perpendicular to the
flow, shows the fluid flow. In practice, time lines often consist of a row of small particles,
such as hydrogen bubbles.
• A streak line arises when dye is injected in the flow from a fixed position. Injecting the dye
for a period of time gives a line of dye in the fluid, from which the fluid flow can be seen.
• A path line is the path of a particle in the fluid. Imagine a light-emitting particle in the flow.
A path line is obtained when a photographic plate is exposed for several seconds.
We will see in section 3 and 4 that time lines, streak lines, and path lines also play an
important role in computer graphics flow visualization. In steady flows, streak lines and path lines
coincide. In that case, streak lines and path lines are identical to stream lines, lines that are
everywhere tangent to the velocity field.
Time lines, streak lines or path lines can be visualized by injecting foreign material in the fluid,
which in general is transparent. As already mentioned, streak lines can be created by injecting dye
in the (liquid) fluid. An example of an application of flow visualization using dye is shown in
figure 2.1. In case of a gas flow, smoke can be injected in the fluid. Though conceptually simple,
these techniques require very careful application. The injection itself can disturb the flow. The
density and temperature of the injected material, if not equal to that of the flow at the injection spot,
can also influence the flow field.
Path lines are generated by adding small particles to the fluid. An example of suitable particles
for liquid fluid is magnesium powder; in gaseous fluids, oil droplets can be used. The velocity can
be measured by photographing the motion of the particles with a known exposure time (see figure
2.2).

3
Figure 2.1 Visualization with dye to study water flow in a river model
(Courtesy Delft Hydraulics)

4
 Figure 2.2 Flow in a harbour model, visualized with particles (Courtesy Delft Hydraulics)

Electrolytic and photochemical techniques can be used to generate time lines. Dye injection
techniques are difficult to apply, because the use of several dye injectors would disturb the flow
significantly. Electrolytic techniques only need a small and thin electrode in the fluid. A well
known technique is the hydrogen bubble technique. This technique is based on the electrolysis of
water. When electrodes are inserted in the fluid and a voltage is applied to the electrodes, hydrogen
bubbles are formed at the cathode and oxygen bubbles are formed at the anode. By isolating some
parts of the electrodes and by varying the voltage level, all kinds of combinations of time and
streak lines can be obtained.
Photochemical techniques can be applied if the fluid should not be disturbed at all. These
techniques also produce a visible tracer (dye) in the fluid by focusing a bundle of light onto a point
in the fluid, or by using a laser to produce a dye along a line in the fluid. The dye is produced by a
photochemical reaction.
Special techniques exist to study the flow field on a surface. One way to visualize flow fields
in the neighbourhood of a surface is to fix tufts (small threads) at several points on the surface.
The orientation of the threads indicates the direction of the velocity. Another way is to coat the
surface with a viscous material such as oil. The fluid flow will produce patterns in the viscous
material, see figure 2.3. These patterns contain information on both the direction and the
magnitude of the fluid velocity.

5
 Figure 2.3 Oil flow pattern visualizes the skin friction distribution of an axisymmetric model
at incidence in a supersonic flow (Courtesy High-Speed Laboratory, Dept. of Aerospace
Engineering, Delft University of Technology)

2 . 2 Optical Techniques
Optical flow visualization techniques are based on the effect that a change in density causes a
change in the light refraction index. These techniques can only be applied when the density of the
fluid is not constant.
There are three basic optical techniques: shadow techniques, schlieren techniques and
interferometry. The simplest technique of these is the shadow technique. Shadowgraphs are
created by passing a parallel light beam through a moving fluid (figure 2.4). The density variations
will cause some of the light rays to be refracted. If the light beams are focused on a photographic
plate, the refraction of some of the light beams results in dark and light areas, see figure 2.5.

photographic
lens lens
film

light
source lens

flow region

Figure 2.4 Schematic arrangement of a shadowgraph system

6
 Figure 2.5 Shadowgraph visualizes the density distribution in the flow around an
axisymmetric model in a supersonic flow (Courtesy High-Speed Laboratory, Dept. of Aerospace
Engineering, Delft University of Technology)

The schlieren technique resembles the shadowgraph technique. It also uses a parallel light
beam, but now two diaphragms are used. One diaphragm (inserted before the flow region) selects
the light beams which are passing through the flow region. A second diaphragm (after the flow
region) stops some of the refracted light beams. This technique is used to visualize density
gradients.
Interferometry is based on the fact that a change in density not only results in a refraction of
the light, but also in a phase shift. In an interferometer parallel light is split into two beams. One of
the beams enters the flow field, the other beam does not enter the flow field. When both beams are
merged and projected on the same photographic plate, interference occurs when the phase of one
of the beams is shifted by a change of density in the fluid flow.

2 . 3 Addition of Heat and Energy

Instead of adding material to a fluid flow, it is possible to add energy to the fluid to visualize its
flow. Especially in cases were the techniques of 2.1 and 2.2 cannot be applied, for example with
low density fluid flows, these techniques are useful.
Heat can be added to a fluid to artificially change the density. If the pressure is constant,
adding heat to a portion of the fluid results in a lower density. The fluid can then be visualized
with techniques as described in 2.2.
Visualizing velocity profiles in gaseous flows is possible with the spark tracing technique.
Two electrodes produce a spark discharge. This discharge ionizes the path of the spark. The light-
emitting ionized fluid elements can be traced through the fluid.
A technique which can be used even for very low density fluid flows is electron-beam flow
visualization. A beam of electrons traverses the gaseous fluid. When electrons collide with gas
molecules, these gas molecules will be excited and emit radiation. The intensity of the radiation is
7
approximately proportional with the density of the fluid. By moving the electron beam, the entire
flow area can be scanned.

3 Computer Graphics Flow Visualization

Experimental flow visualization is a starting point for flow visualization using computer graphics.
The process of computer visualization is described in general, and applied to CFD (3.1). The heart
of the process is the translation of physical to visual variables. Fluid mechanics theory and practice
help to identify a set of ‘standard’ forms of visualization (3.2). To prepare the flow data to be cast
in visual form, several types of operations may have to be performed on the data (3.3). A special
type of operation, analysis of the topology of a fluid flow data set, is described separately (3.4).

3 . 1 The Flow Visualization Process

Scientific visualization with computer-generated images can be generally conceived as a three-stage
pipeline process (Haber and McNabb ‘90). We will use an extended version of this process model
here (see figure 3.1), and discuss its application to flow data visualization.

DATA GENERATION

DATA ENRICHMENT
& ENHANCEMENT (3.3)

VISUALIZATION MAPPING
(3.2)

RENDERING (4)

DISPLAY (4.3.1)

Figure 3.1 A pipeline model of the visualization process

• Datageneration: production of numerical data by measurement or numerical simulations.
Flow data can be based on flow measurements, or can be derived from analysis of images
obtained with experimental visualization techniques as described in section 2, using image
processing (Yang ‘89). Numerical flow simulations often produce velocity fields,
sometimes combined with scalar data such as pressure, temperature, or density. In the rest
of this paper, we will be mainly concerned with data generated by CFD simulations.
• Data enrichment and enhancement: modification or selection of the data, to reduce the
amount or improve the information content of the data. Examples are domain transforma-
tions, sectioning, thinning, interpolation, sampling, and noise filtering. We will discuss
some of these operations in 3.3.

8
 • Visualization mapping: translation of the physical data to suitable visual primitives and
attributes. This is the central part of the process; the conceptual mapping involves the
‘design’ of a visualization: to determine what we want to see, and how to visualize it.
Abstract physical quantities are cast into a visual domain of shapes, light, colour, and other
optical properties. Some ‘standard’ conceptual visualization mappings of flow data will be
discussed in 3.2.
The actual mapping is carried out by computing derived quantities from the data suitable for
direct visualization. For flow visualization, an example of this is the computation of particle
paths from a velocity field. As this type of operations is closely related to the operations of
the enrichment and enhancement stage, we will discuss these together in 3.3 under the
heading Data Preparation.
• Rendering: transformation of the mapped data into displayable images. Typical operations
here are viewing transformations, lighting calculations, hidden surface removal, scan
conversion, and filtering (anti-aliasing and motion blur). Rendering techniques for flow
visualization will be discussed in section 4.
• Display: showing the rendered images on a screen. A display can be direct output from the
rendering process, or be simply achieved by loading a pixel file into a frame buffer; it may
also involve other operations such as image file format translation, data (de)compression,
and colour map manipulations. For animation, a series of precomputed rendered images
may be loaded into main memory of a workstation, and displayed using a simple playback
program.

3 . 2 Flow Visualization Mappings

The style of visualization using numerically generated data is suggested by the formalism
underlying the numerical simulations. The two different analytical formulations of flow: Eulerian
and Lagrangian, can be used to distinguish two classes of visualization styles:
• Eulerian: physical quantities are specified in fixed locations in a 3D field. Visualization
tends to produce static images of a whole study area. A typical Eulerian visualization is an
arrow plot, showing flow direction arrows at all grid points.
• Lagrangian: physical quantities are linked to small particles moving with the flow through
an area, and are given as a function of starting position and time. Visualization often leads
to dynamic images (animations) of moving particles, showing only local information in the
areas where particles move.
Other styles may be called indirect visualization of flow fields. A velocity field can be charac-
terized by deriving certain scalar quantities, which can be visualized using volume rendering
techniques (see 4.2.6). Examples are the magnitude of velocity or vorticity, or helicity (Buning
‘89). Also, instead of individual particles, concentration of particles per volume unit can be
computed, and visualized as a scalar concentration field (Hin et al. ‘90). A second type of indirect
visualization shows the effect of the flow on the geometry of certain objects, eg. deformation of
spheres (Geiben and Rumpf ‘91).
There is a number of visual representations, that can be conceived as ‘standard’ visualization
mappings in the sense of the process model described in 3.1, some of which are directly based on
experimental flow visualization: velocity arrows, stream lines, stream surfaces (ribbons), streak
lines, path lines, time lines (surfaces), and contours. They can be visualized using obvious graphi-
cal primitives such as points, lines, and surfaces. In section 4, we will also discuss some more
advanced presentation techniques
Arrows are icons for vectors, and the most obvious way to visualize a velocity field with
computer graphics is an arrow plot. Usually, arrows are tied to grid points, and indicate both

9
direction and speed at these points. In this way, a complete view of a vector field can be given; this
is a clear example of an Eulerian style visualization. Used in this way, arrows can give reasonable
results in 2D, but there are numerous problems in the 3D case. We will discuss some of these
problems in 4.2. A special type of arrow is the stream vector, where the tail of the arrow follows a
stream line.
Stream lines, streak lines, path lines (or particletraces), and time lines are directly derived
from experimental visualization (see 2.1). They can be generated and rendered as 2D or 3D curves
(see figures 3.2, 3.3, 3.4). In 3.3, some techniques used for computation of these curves will be
discussed; we will return to rendering in 4.2 and 4.3.
In 2D, stream lines characterize a general directional flow pattern very well (figure 3.2). Also,
they give an indirect cue to velocity magnitude, because the distance between stream lines is
inversely proportional to velocity.

Figure 3.2 Stream lines in 2D (Bertrand and Tanguy ‘88)

In 3D, stream lines can also be used, but additional depth cues are needed to locate the curves
in space. One possibility is to generate a stream surface, interpolating a set of adjacent stream lines
(Hultquist ‘90). A surface interpolating only two adjacent stream lines is called a stream ribbon
(Buning ‘89, Hultquist '90).
Streak lines and path lines (see figure 3.3) can be visualized as static curves, and this can be
interpreted either as a time interval in a steady flow, or as a single instant in an unsteady flow.
Animation can show the curves dynamically as they change over time. Particles can also be
visualized as moving dots or objects, or as a moving texture (Van Wijk ‘90b, ‘91; see 4.2.4 and
4.3.5).

10
 Figure 3.3 Particle traces (Strid et al. ‘89)

Time lines drawn for a series of instants clearly show the distribution of velocity (figure 3.4).
When a 2D array of particles is released at one instant in a 3D flow field, they can define a time
surface at each instant.

Figure 3.4 Time lines

Contours are curves (in 2D) or surfaces (in 3D) where a given scalar physical variable has a
constant value. Contours, also called iso-curves and iso-surfaces, do not have counterparts in
experimental visualization. Their applicability is limited to scalar data, such as pressure,
temperature, or velocity magnitude, or scalar quantities derived from a vector field. An example of
the latter is the display of pressure contours on the surface of an airplane fuselage (see figure 3.5).

Figure 3.5 Pressure contours on a surface

In 3D, contour surfaces have the advantage of visualization with a full range of visual depth
cues, but a problem is the display of multiple contours in one picture. Dynamic probing of a scalar
field by interactively varying the contour value, and showing a changing contour surface is a good

11
solution. This is hard to perform in real time, but this may be simulated by showing sets of
precomputed contour surfaces; other forms of pre-processing, such as span filtering (Gallagher
‘91) can also be used to speed up this process.

3 . 3 Data Preparation
Visualization usually starts with ‘raw data’ resulting from the data generation process, and often
these data are not suitable for direct visualization in the given form. Data preparation consists of a
variety of operations, including data enrichment and enhancement, and computation of derived
quantities for a certain visualization mapping.
We will give some examples of operations performed at this stage. Some operations are appli-
cable to any type of data, and some are based on the presence of a grid, in which data values are
given for the node positions.
3.3.1 Filtering
Measured data will always contain noise and outlyers, or peak values, which may disturb
visualization. The data can be viewed as samples from a continuous signal, and the image as a
reconstruction of that signal. In terms of signal processing, the source signal may contain too
many high frequency components, caused by measurement noise and peaks. Filtering can be
applied to remove these spurious high frequencies.
3.3.2 Data Selection
To reduce the amount of data to be visualized, and to concentrate on the most interesting parts or
features of the data, data selection techniques must be applied.
The simplest selection is a global thinning by sub-sampling, or by aggregation and averaging
of the whole data field. Also, a part may be cut out by clipping the data against a given volume.
More sophisticated selection can be done by calculating some ‘interestingness’ index for each grid
cell, and only visualizing cells with a high value of this index. Measures for this may be local
extreme values of a quantity, or large gradients, such as sudden changes in velocity. The
computed indices can be treated as a scalar field, and volume rendering may be applied for
visualization. Thresholding, or more advanced image processing techniques can be used to
decompose the data in meaningful parts.
A last group of techniques is the extraction of specific flow features or patterns, such as flow
field topology or vortex cores (Helman and Hesselink ‘90; Globus et al. ‘91); see section 3.4.
3.3.3 Domain Transformations
The process of numerical simulation and visualization of fluid flow is typically performed in three
different domains:
• the physical domain P: here, the equations of motion are defined; the domain is discretized,
often into a curvilinear, boundary-conforming grid, fitting the surface of objects; the flow
variables (velocity, density, pressure, etc.) are computed in the grid points of P.
• the computationaldomain C: the equations of motion are transformed to this domain; it is
discretized to suit the needs of numerical computation, often into a uniform rectangular
grid, and thus deformed with respect to P.
• the graphicaldomain G: often also discretized, to suit the needs of graphics processing.
There is no generally accepted representation of G. As visualization often directly refers to
physical reality, the shape of objects in G is usually the same as in P. Often a regular or
hierarchical, rectangular discretization is used. G is populated with geometric primitives
and attributes, such as shapes and colours, which must be ultimately expressed in pixels.

12
 x = L(x')
v = Jv'
ζ

η
(i, j, k) ξ (i, j, k)
x y x' = L-1 (x)
v' = J -1 v computational space (ξ,η,ζ)
physical space (x,y,z)

Figure 3.6 Transition between physical space P and computational space C.

Transitions between these domains must be performed, often in both directions (see figure
3.6). The first step is discretization or grid generation, as a part of the pre-processing for
numerical computations. Grids can be of several types: structured or unstructured, rectilinear or
curvilinear, and combinations of these (Speray and Kennon ‘90). Even more complex situations
may occur when grids are staggered, which means that data values (such as velocity vector
components) are computed in different faces of each grid cell. We will restrict the discussion here
to structured grids, with regular hexahedral topology; the geometry of the grids can be curvilinear,
resulting in cells with a warped brick shape, or orthogonal, resulting in cells with a cubical shape
(Cartesian grid) or a rectangular brick shape.
The discretization in P for a flow simulation often is the definition of a structured, curvilinear
grid with cell indices (i,j,k), such that the grid point nearest to the origin of the grid has
coordinates (i,j,k). A general point of P is denoted as x p (x,y,z). Velocity vectors v p (i,j,k) =
(u,v,w) are computed at each grid point.
Computational space C is usually discretized as a regular orthogonal Cartesian grid with the
same cell indices (i,j,k), and points x c (ξ,η,ζ). Velocities at the grid points of C are v c (i,j,k) =
(u’,v’,w’). Generally, a single global transformation between P and C is not known, but for each
neighbourhood of a grid point (i,j,k) in P a local transformation L can be determined.
Transformations for other points in P must then be derived by interpolation between grid values.
L specifies the local transformation of a grid point (i,j,k) from C to P as x p = L(x c); similarly,
L-1 is used to transform a point from P to C (see figure 3.6). The Jacobian matrix J of L, defined
analytically as J = ∂L/∂x c, can be used to transform a vector quantity from C to P. For instance,
v p = J·v c. Again, the inverse J-1 is used to transform a vector from P to C.
In general, the mappings are only known at discrete points. As a consequence, the Jacobian
must be approximated by finite differences. For a grid point (i,j,k) of C, the columns of J may,
for example, be approximated by:
Je 1 = x i+1,j,k – x i,j,k (3.1)
Je 2 = x i,j+1,k – x i,j,k (3.2)
Je 3 = x i,j,k+1 – x i,j,k (3.3)
where x i,j,k are the coordinates of grid point (i,j,k) in C, and e i the unit vector in the direction
1
of xi. Another possibility would be to use central differences: Je 1 = 2
(x i+1,j,k –x i-1,j,k ); other
types of differences can also be used.

13
 As visualization often directly refers to physical reality, G must be undeformed with respect to
P. Also, a new discretization is desirable in G, to support the operations in the rendering stage.
The transition to another grid usually involves a resampling of the data field, and this has several
disadvantages. Especially the transition from a boundary-conforming, locally refined curvilinear
grid in P to a uniform orthogonal grid in G may lead to a severe waste of storage space, or to loss
of information, depending on the resolution of the regular grid. In areas where the resolution of
the P grid is higher than the G grid, data may be lost, while in low resolution areas of P,
oversampling will lead to many identical data points in G. A partial solution is the use in G of a
hierarchical grid type of which the resolution can vary locally, such as the octree (Samet ‘90).
Often, the boundary conformance will be lost, so that object geometry must be represented
separately in G. Another important point is the degradation of accuracy as a result of interpolation.
Use of higher order interpolation techniques can reduce the loss of accuracy.
3.3.4 Interpolation
Flow quantities are usually given only at grid points, so that for other points values must be
obtained by interpolation. Interpolations may be of zero, first, or higher order, depending on the
accuracy required. In the following, we will assume a regular orthogonal grid (as often used in
computational space C), in which the cells are cubes with (i,j,k) as integer indices; α , β, and γ are
the fractional (0…1) offsets of a point within a cell; E(i,j,k) = Eijk is a grid point value.
For zero order interpolation, the value within each grid cell is assumed to be constant. The
value for a point inside the cell is either the average of the surrounding eight grid points, or equal
to the value in the nearest grid point. This interpolation is discontinuous over the grid cell
boundaries.
In first order interpolation, a linear variation of the value in the grid cells is assumed. With tri-
linear interpolation, a linear interpolation is performed in x-, y- , and z-direction using the fractions
α , β, and γ , respectively (see figure 3.7).

1–γ

1–β
z y γ
β
x α 1–α
Figure 3.7 Tri-linear interpolation; interpolation in the x-direction gives four points in the
plane x=α , interpolation in the y-direction gives two points on the line x=α , y=β, interpolation in
the z-direction gives the interpolated value at (α , β, γ )

In a regular orthogonal grid, α , β, and γ can be easily determined from the coordinates. The
interpolation can then be described for a single cell (with i,j,k = 0,1), with two linear basis
functions: Φ 0(s) = 1–s, and Φ 1(s) = s; the product of three basis functions is defined on each of
the eight corner grid points of the cell:
Φ ijk (α , β, γ ) = Φ i (α )⋅Φ j (β)⋅Φ k (γ ), with: i,j,k = 0 or 1. (3.4)

14
 Now the interpolated value is obtained as:

Ε αβγ = ∑ Eijk⋅ Φijk(α , β, γ ) (3.5)

i,j,k=0,1

Higher order interpolations work in a similar way, using higher order basis functions. The
order of continuity of the interpolated values across grid cells is lower than the order of the
interpolation. Thus, C0 (positional) continuity can be obtained with first order interpolation, and
for C1 (tangent) continuity at least second order interpolation is needed.
Interpolation in a regular curvilinear grid, as is often the case in physical space P, is much
more complex. The fractions (α , β, γ ) are unknown in this case, and must be computed. We can
use the same interpolation method described above; starting from a given point x p (x,y,z) in cell
(i,j,k), we have:

xp = ∑ x ijk⋅ Φijk(α , β, γ ), (3.6)

i, j, k=0,1

giving three non-linear equations, which can be solved numerically to find α , β, and γ .
Alternatively, α , β, and γ can be determined using the method described below.
3.3.5 Point Location in Computational and Physical Space
It is often necessary to determine the location of a point with respect to a grid, that is, given the
coordinates of a point, we must find the grid cell indices (i,j,k), and fractional offsets (α , β, γ )
within the cell containing that point. If we have a point x c (ξ, η, ζ) in C, it is easy to find the cell, if
the grid in C is regular and orthogonal.
If we have a point x p (x,y,z) in a structured curvilinear grid in P, and we must find the corre-
sponding point in x c in C, an incremental search through the cells must be performed, starting
from an initial point, each time taking a step towards a neighbouring cell. One algorithm for this
search is the stencil walk (Buning ‘89).
We have a point x p in P, and we want to find the corresponding cell (i,j,k) and offsets (α , β, γ )
in C. An initial guess for the right cell in P can be made by finding the nearest grid point, and
choosing one of the cells adjacent to that grid point. In the corresponding initial cell in C, we
choose x c*, with α =β=γ =0.5, the center of the cell. This point is transformed to x p*in P, deter-
mined by tri-linear interpolation between the P-coordinates of the corner points of the cell; for the
center point, this amounts to taking the average.
The difference vector ∆x p = x p*-x p is again transformed from P to C, using the inverse
Jacobian matrix J-1, which is found by interpolation between the values of the corner points of the
cell, again with α =β=γ =0.5. In C, we thus find a difference vector ∆x c = (∆α ,∆β,∆γ ) = J-1 ∆x p,
which is used to determine new values for α , β, and γ by adding the old values (0.5) to ∆α , ∆β,
and ∆γ .
If at least one of the new values (α , β, γ ) is outside [0,1], the point is not in the current cell, and
the search continues. If α > 1, then i is incremented, and if α < 0, i is decremented; similar rules
apply for β and γ . The center of the cell is used as the new starting point, and this continues until
(α , β, γ ) are within [0,1], which means we have found the right cell. Then we can continue in the
same way within this cell, until |∆x c| is small enough.
In cases when a grid contains holes or cavities, this procedure may fail when the search path
crosses a boundary of the grid. This must be detected, and the search can be continued by
following boundary cells to get around the obstacle.

15
 Point location is greatly simplified if a previous location of the point is known, as often
happens with particle tracing. In that case, the cell of the previous location (or a neighbouring cell
in the direction of v ) can be used as an initial guess. In an unstructured grid, it is necessary for this
purpose to store adjacency information of the cells.
3.3.6 Computing Derived Scalar Quantities
Several scalar quantities can be computed from a velocity field for each grid cell, and may be used
to indirectly visualize the flow field. We give only a few examples here.
• The magnitude of all velocity vectors ||v || defines a scalar field.
• The kinetic energy density is 12 ρ⋅ ||v ||2 .
• The scalar product of two vectors is a measure of the angle ϕ between them:
u⋅ v = || u || || v || cos ϕ (3.7)
This can be used to find the components of all velocity vectors in a given direction, or to
find the changes in direction at two neighbouring points.
• The magnitude of the vorticity ω (see 3.3.7) may be used to find vortices. Using ω , helicity
density is computed as: Hd = v ⋅ ω . (Buning ‘89)
• The scalar fshock is defined for compressible media as:
∇p v
fshock = ⋅ , (3.8)
|∇p| c
in which c is the speed of sound. The iso-surface for fshock = 1 shows shock waves
(Buning ‘89).
3.3.7 Computing Particle Path Lines and Other Vector Quantities
Particle path lines, or integral curves, can be computed from the velocity field by computing a
series of consecutive particle positions in the field, and fitting a curve through them. The motion of
a particle is governed by the equation:
dx
dt = v (x), (3.9)

with x the position vector of a particle at a given instant t, and v (x) the velocity field.
Integration of this equation yields position x at any instant. As the velocity is usually only given at
the grid point positions, the velocity at other positions must be obtained by interpolation.
The problem can thus be stated as determination of particle positions in physical space P,
starting from an initial position x (x,y,z). This is performed in computational space, in the follow-
ing three steps (Buning ‘89):
1. find the position of x in computational space C, as a grid cell (i,j,k) and offsets (α , β, γ ) in
the cell;
2. interpolate within the grid cell to find the velocity vector at that point;
3. integrate equation (3.9) to find the next particle position.
The first two steps have already been discussed above (3.3.5 and 3.3.4). For the third step,
several integration schemes can be used; each scheme gives an approximation of:

x(t+∆t) = x(t)+ ∫ v (x(t))dt, (3.10)

for a time step ∆t. The simplest is the first order Euler technique. The integral is approximated
as v (x(t)) ∆t, so that x(t+∆t) = x(t) + v (x(t)) ∆t. This approximation is too inaccurate; a second
order Runge Kutta technique (also known as the Heun scheme) gives better results. Besides the

16
velocity at x at instant t, we also use an estimation of the position x * at instant t+∆t; for this
predictor step, we use the Euler method:
x * (t+∆t) = x(t) + v (x(t)) ∆t. (3.11)
Now x * (t+∆t) is used to estimate velocity at t+∆t, and x(t+∆t) is computed in a corrector step:

x(t+∆t) = x(t) + 12 [v (x(t)) + v (x * (t+∆t))]⋅ ∆t (3.12)

The choice of the time step ∆t also affects accuracy. The optimum time step size is a
compromise between accuracy and computational cost. Use of a variable time step, depending on
the gradients in the velocity field, is the best solution. This may be done eg. with ∆t = α /v a, where
α is the number of steps per cell, and v a is the average velocity of the eight surrounding grid
points.
For visualization of moving particles using animation, particle positions must be known at
equal time intervals for each frame, to show velocity information. If particle positions were
computed with variable time steps, each particle path must be resampled to find positions at equal
intervals.
If we are able to compute particle traces, other objects can be easily derived. Stream surfaces
(ribbons) can be constructed by fitting surfaces through particle positions, using polygons, or any
type of smooth surface interpolation. Streak lines can be simulated by releasing a continuous flow
of particles from a single point. Time lines (-surfaces) can be obtained by releasing a 1D (2D)
array of particles at one instant t, and interpolating a curve (surface) through the positions of these
particles at any given instant.
In a 3D incompressible flow, stream lines are curves that satisfy the equation:
dx dy dz
u = v = w, (3.13)

where u, v, and w are the velocity components in x-, y-, and z-direction. For a time-
independent flow field, the calculation of stream lines amounts to the same as computing particle
paths (Strid et al.’89).
Vorticity is another important vector quantity that can be computed from the velocity field. For
this computation, the curl of the velocity field must be determined, since vorticity is defined as ω =
∇ × v.

3.3.8 Contour Lines and Surfaces

Contour lines can be considered in 2D as intersection curves of a surface defined by a scalar
function z = f(x,y) and a (horizontal) plane z = c. Usually, no scalar function is known, but a field
is defined by scalar data values at grid points. To determine contour lines, each grid cell is
examined; if grid values both higher and lower than the contour value are found in a cell, it
contains a part of the contour line. Intersection points can be determined with the edges of the cell,
by linear or non-linear interpolation, and these points are connected by line segments, or a smooth
curve is fitted through them. There are several methods to generate contour lines, for different
types of grids, interpolations, and orders of curve generation. For an overview, see Sabin (‘85).
In the case of a 3D scalar field, a stack of planes, each with contour lines, may be determined
for a number of positions in the i-, j- or k-direction of a regular rectangular grid. This can give
some suggestion of 3D contour surfaces, but for properly shaded display other techniques must be
applied. In volume rendering, contours are rendered directly by selective display of voxel values

17
above a certain threshold; surface normals for the lighting calculations are estimated from gradients
(Levoy ‘88). This technique gives visually good results, without explicit calculation of the contour
surface.
Another technique often applied in volume visualization is the Marching Cubes algorithm
(Lorensen and Cline ‘87). This algorithm finds an explicit polygonal representation of a contour
surface. The principle is as follows:

mark all cells as undone

find a starting cell, and put it into a queue
while the queue is not empty do
get the next cell from the queue
process the cell as described below
mark the cell as done
for each of the intersected edges:
put all unmarked neighbour cells sharing the edge into the queue

In a structured grid, the edge-sharing neighbours can be easily found with the cell indices; in
an unstructured grid, adjacency information is needed.
For a hexahedral (cubical) cell, all points on the edges with the contour value are determined
by linear interpolation. As there can only be one point per edge, at most 12 points can be found in
a cubical cell. Now the intersection of the cell with the contour surface is determined, consisting of
one to four triangles. This is done by encoding the eight vertices of the cell as higher than, or
lower than the contour value. The resulting 256 possible different binary code values were reduced
to 14 different cases, each corresponding with a distinct configuration of triangles. The binary
code for each cell is used as an index into a table of configurations, which is used to generate the
output triangles, with the edge intersection points as vertices.

3 . 4 Flow Field Topology

Since 1987 the analysis and visualization of flow topology has been investigated thoroughly. In a
number of papers, Helman and Hesselink ('87, '88, '89abc, '90, '91) describe the development
of a system to visualize flow topology. A general classification of flow fields is described by
Chong et al. ('90). More recently, similar techniques to those of Helman and Hesselink were
described. Dickinson ('91) addresses the interactive aspect of flow topology visualization, and
Globus et al. ('91) give detailed information on how to implement a flow topology visualization
module.
Flow topology analysis is based on critical point theory, which has been used widely to
examine solution trajectories of ordinary differential equations. The topology of a vector field
consists of critical points (where the velocity vector is zero) and integral curves and surfaces
connecting these critical points. Images of a vector field topology display the topological
characteristics of a vector field, without displaying too much redundant information.
The following steps are necessary to analyze and visualize vector field topologies:
• the location of the critical points must be calculated
• the critical points must be classified
• integral curves and surfaces must be calculated
We will discuss these steps in more detail below. It should be noted that, although the
techniques will be described for velocity fields, they can also be applied on any other vector field,
such as vorticity fields or pressure gradient fields.

18
3.4.1 Critical points
The positions of the critical points can be found by searching all cells in the flow field. Critical
points can only occur in cells where all three (or two, in 2D) components of the vector pass
through zero. The exact position of a critical point can be calculated by interpolation in case of a
rectangular grid. In case of a curvilinear grid, the position of a critical point can be calculated by
recursively subdividing the cell, or by a numerical method such as Newton iteration (Globus et al.
‘91).
Once the critical points have been found, they can be classified. This is done by approximating
the velocity field in the neighbourhood of the critical point x cp with a first order Taylor expansion.
This gives the following formula for the velocity u:
∂u
ui ≈ ucp , i + (x j - x cp , j ) ∂x i . (3.14)
j

Because the velocity in a critical point is zero, the velocity field in the neighbourhood of a
critical point is fully determined by the partial derivatives ∇u = ∂ui /∂x j . The critical points can be
classified according to the eigenvalues and eigenvectors of ∇u. Figure 3.8 shows some of the
configurations for a three-dimensional vector field. In these figures, positive eigenvalues
correspond to velocities away from the critical point (repelling nodes), and negative eigenvalues
correspond to velocities towards the critical point (attracting nodes). Complex eigenvalues result in
a focus; if the real part is non-zero, a spiral occurs, and if the real part is zero, concentric ellipses
occur. If both negative and positive real values exist at a critical point, the critical point is a saddle.

a) b) c)

e) f)
d)

Figure 3.8 Examples of three dimensional critical points a) repelling focus, also repelling in
the third dimension, b) repelling node, c) saddle, repelling in the third dimension, d) attracting
focus, repelling in third dimension, e) attracting node, f) center, repelling in the third dimension

3.4.2 Integral curves and surfaces

The classified critical points can be used as starting points for integral curves (‘particle paths’), and
the eigenvectors can be used as starting directions. This means that the starting point is a point on
an eigenvector, very close to the critical point. The end points of the integral curves are critical
points again, or are points on the boundary of the flow domain. Because of numerical errors,
some integral curves might ‘miss’ a critical point, or might enter an object in the flow. By
attaching an integral curve to a critical point when it comes very close to the critical point, the first

19
problem can be solved. To solve the second problem, integral curves that enter an object can be
restricted to follow the surface of the object.
In two dimensions, critical points and integral curves completely describe the flow field in a
qualitative way. In three dimensions, integral surfaces should also be calculated. Helman and
Hesselink ('91) describe a way to create some of these integral surfaces. When visualizing the
flow around a hemisphere cylinder, they first calculate the topology of the vector field on the
surface of the cylinder. This is, in fact, a two-dimensional topology. Points on the (two-
dimensional) integral curves of this topology are used as starting points for a number of three-
dimensional integral curves. By tessellating the space between two adjacent integral curves,
integral surfaces can be visualized. In this way they are able to show surfaces of separation and
reattachment.

4 Presentation Techniques
After data preparation and visualization mapping, the flow data have been cast into a form suitable
for visual presentation. The visualization objects can be transformed into pictures, together with
additional objects such as the environment of the flow, and auxiliary objects such as scales,
pointers, colour bars, and numerical or textual annotations. First, we will turn to some aspects of
human perception, and relevant visual cues to achieve optimum results for 3D flow visualization.
Then we will discuss basic rendering techniques in 4.2, and some more advanced methods in 4.3.

4 . 1 Human Perception and Depth Cues

In 3D flow data visualization, often complex spatial structures must be shown, and for this a good
orientation in 3-space is essential. Human visual perception can readily extract spatial information
from a 2D picture, provided that enough supporting ‘depth cues’ are available. Many depth cues
are related to objects and surfaces, but field data (such as scalar and vector fields) do by nature not
contain objects or surfaces and thus lack these cues. For visualization, extra depth cues must then
be added, 3D objects and surfaces derived from the fields must be used, or environment geometry
must be exploited to support localization in space.
We will give some examples of important cues for the perception of shapes, depth, and
motion. (For a good introduction and further references, see Thorpe ‘90.)
3D shapes are mainly perceived by their contours, and the directional reflection of light on
surfaces, showing as changes in reflected intensity. Gradual changes indicate a smoothly curved
surface, while discontinuous changes indicate edges.
Depth or spatial information is perceived in many ways:
• perspective: parallel edges converge into the distance; the size of objects decreases linearly
with their distance from the observer;
• stereoscopy: the disparity between images seen with two eyes gives a powerful depth
effect, which is independent from other depth cues, and can be applied to virtually every
type of image;
• occlusion: nearer objects cover other, more distant objects;
• surface texture density gradients: a continuous density change implies a surface expanding
into depth, a discontinuity is perceived as a contour or an edge;
• shadows cast by one object onto the ground or onto other objects give information on
spatial relationships;
• distance cues, such as atmospheric haze: the diminishing of colour contrast in the distance;
• artificial optical cues, such as depth of field;
20
 • user control: if a user is given the possibility to navigate around the 3D display space, to
probe and interrogate the data, this will help the user to build a 3D mental image of the data
space. An example from flow visualization is the interactive positioning of particle sources
(Wavefront Technologies ‘90), and the virtual wind tunnel (Bryson and Levit ‘91), where
the user can move around in 3D data space, and can use data glove gestures to generate
particle path lines in real time.
Depth ambiguity can easily occur when spatial cues are absent or too weak. This is quite
common in line drawings (such as arrow plots), which contain no surface information. To support
localization of an object floating in space, it can be related to the environment, or to coordinate
planes by showing the object’s coordinates or projection on two or three planes.
Motion can also support depth information. Changes in viewpoint resulting in gradual changes
of occlusion relations and perspective, are known as motionparallax; this is readily perceived as
extra depth information. The world as perceived by a moving observer has the characteristics of an
optic flow field (Gibson ‘50,’79). A static world ‘flows’ around a moving observer in a standard
pattern. If an object is perceived which moves in a way that is different from this pattern, it is
detected to be in motion.
Moving objects can show direction and velocity information. A dense cloud of small objects
can be perceived as a moving texture (Upson et al. ‘89; Van Wijk ‘90b, ‘91), and can be given
certain surface properties to improve localization in space. To preserve velocity information of
moving objects (particles), the screen update time must be proportional to a simulation time step; in
practice this means that both must be constant. Also, certain precautions must be taken to reduce
the disturbing and possibly false effects caused by discretization (see 4.3).
In flow visualization, knowledge of human perception can be used to improve communicaton
of spatial and motion information. Below we will show some examples where shape, depth, and
motion cues are consciously applied for this purpose.

4 . 2 Basic Rendering Techniques

In the visualization mapping stage, visual primitives have been defined to represent the data.
Generally speaking, the rendering process consists of four stages: viewing transformations,
visible surface determination, lighting calculations, and scan conversion. We will not attempt to
summarize the basic rendering techniques for these primitives. Many techniques for rendering
lines, curves, polygons and curved surfaces can be found in any computer graphics textbook (eg.
Foley et al. ‘90). Rendering solids is described in Bronsvoort et al. (‘91), and volume rendering in
Kaufman (‘90). Often, a visualization will consist of a combination of several types of primitives,
and therefore it is profitable if a visualization system permits the use of many different primitives
in a single image.
We will concentrate here on the visualization mappings listed in 3.2, and the use of extra depth
cues to improve spatial representation of 3D flow fields.
4.2.1 Arrows
Arrows can simply be drawn using only straight lines. In 2D, this can give reasonable results,
provided that the arrows are scaled so they do not overlap (figure 4.1).

21
 Figure 4.1 2D arrow plot (Bertrand and Tanguy ‘88)

In 3D, rendering arrows is much more complex. Perspective is the only depth cue available; there
is no occlusion, or directional light reflection to assist depth perception. 3D line arrows are
ambiguous in direction; for example, it is impossible to distinguish between arrows pointing
towards and away from the observer (figure 4.2).

Figure 4.2 Directional ambiguity of 3D arrows

The size of the arrows is determined by three factors: velocity magnitude, direction with respect to
the image plane, and perspective. Also, the display of a full 3D array of arrows very soon
becomes cluttered. Various types of arrows have been devised to minimize cluttering, and to
improve the directional effect (Kroos ‘84), but none of these can fully solve these problems.
An improvement in this respect is to relate the arrows to a plane (figure 4.3), and to show
projections (‘shadows’) of the arrows on the plane. The effect is similar to the use of tufts in
experimental visualization.

22
 Figure 4.3 Arrows connected to planes, with projections

The flow field can be sliced, moving a section plane through the field, in the same style as
applied in volume rendering. Observing the gradual changes of the arrows on the plane allows the
user to mentally imagine the entire field.
More spatial cues can be added by drawing arrows as 3D polygonal objects. The occlusion in
a visible surface display gives a good depth cue, and reduces directional ambiguity. Also,
directional light reflection and shading can be applied, giving some extra information on
orientation in space (figure 4.4). But as polygonal arrows are larger, they can lead to an increase
of cluttering.

Figure 4.4 Arrow as 3D object

4.2.2 Curves
Curves, such as stream lines, streak lines, path lines, and contour lines, can be visualized as a
sequence of short line segments. Again, this gives good results in 2D, but in 3D spatial curves are
hard to localize without further depth cues. Also, only a small number of curves can be displayed
without confusion.
An improvement is showing projections of a curve in the main coordinate planes, allowing
mental reconstruction of the 3D shape by the trained observer (figure 4.5).

23
 Figure 4.5 Spatial curves with projection in a coordinate plane

Another possibility is to display curves as 3D pipes, allowing occlusion and directional light
reflection. (figure 4.6).

Figure 4.6 Spatial curves rendered as pipes (Geiben and Rumpf ‘91)

It is often desirable to display curves (for example iso-curves or stream lines) on surfaces. If
the curves are drawn as lines, their relation with the surface remains weak. A much better result is
obtained when the lines are modelled as 3D strips of constant width pasted onto the surface (Van
Wijk ‘90a). The strips can be rendered transparently, using texture mapping (figure 4.7)

24
 Figure 4.7 Contour lines on a curved surface (Van Wijk ‘90a)

4.2.3 Surfaces
A good directional effect for stream lines is achieved by rendering ribbons (Hultquist ‘90),
which can be produced by connecting two adjacent curves with a maze of polygons. Rendered
with proper lighting and shading, stream ribbons give a good impression of flow direction, also
showing significant effects such as twist and divergence (figure 4.8).

Figure 4.8 Stream ribbons, showing twist and divergence (Hultquist ‘90)

Surfaces, such as stream surfaces (Hultquist ‘90), time surfaces, or iso-surfaces, can be
represented as explicit geometric objects (see 3.3.8), or displayed directly from the data. In
volume rendering, iso-surfaces can be displayed directly using segmentation techniques (Levoy
‘88). Surface normals for shading are approximated by gradients of scalar values. In flow
visualization, surfaces are often defined by points, eg. particle positions. An explicit surface can
be obtained by interpolating a set of points, generating a polygonal mesh or a smooth surface. In
4.3.5 techniques will be described for direct display of surfaces in flow visualization.

25
 Good shape and depth effects can be achieved by many techniques well known in computer
graphics. Visible surface display must be applied, and directed lighting and shading with diffuse
and specular reflections (see Foley et al. ‘90).
A surface can also be used to display extra data pertaining to it, such as pressure and
temperature. The object colour can be varied according to a scalar quantity, and contour lines can
be shown on the surface (Van Wijk ‘90a; see figure 4.7). Texture mapping allows display of
directional information on the surface (Van Wijk ‘91; see 4.3.3). Finally, surfaces can be rendered
semi-transparently, to allow multiple layers of surfaces to be visualized at the same time (figure
4.9). A maximum of three layers can be viewed distinctly in this way.

Figure 4.9 Transparent contour surfaces (Gallagher ‘89)

A special use of a surface in flow visualization is the stream polygon (Schroeder et al. ‘91).
This is a regular, n-sided polygon, positioned in a flow field, oriented normal to the flow
direction. The polygon can be scaled, sheared, and rotated in response to local strain and rotation
in the field, or to other quantities. By sweeping the polygon along a stream line, an n-sided
cylindrical stream tube is obtained. The effects of deformation and sweeping can be combined,
yielding a complex cylinder that is twisted and tapered. The surfaces of the polygon and the stream
tube can be shaded using directed lighting, and can also show associated data. Figure 4.10 shows
an example of this technique.

Figure 4.10 A stream line, ribbon, and stream tube (Schroeder et al. ‘91)

26
4.2.4 Particles
A particle may be rendered as a small-sized object, as a single point, or as a special type of
primitive that has some characteristics of both. As an object, a circle (in 2D) or a sphere (in 3D)
would be an obvious choice. But these objects do not show any changes in orientation. Even with
directed light reflection, a sphere only shows its shape, not its position and orientation in space.
To take advantage of changing light reflection as a spatial cue, a flat or elongated shape should be
used.
To improve the directional effect, especially in still images, a tail can be attached to a particle.
This can be the motion path in the preceding time interval, with the length indicating its velocity
(figure 4.11).

Figure 4.11 Particles with tails (Hibbard ‘89)

A related way to render a particle is the stream arrow. The head of the arrow points to the
particle’s position, and indicates its direction of motion. The tail again is a part of the motion path
(figure 4.12).

Figure 4.12 Stream arrow

Particles can also be rendered as points. This is of special interest to visualize the global
structure of a flow field, using large numbers of particles. Although a point seems the simplest of

27
all geometric primitives, rendering points is by no means trivial (Van Wijk ‘92). If a point is
simply rendered by setting the pixel onto which the point is projected, regardless of its position
within the pixel, serious errors occur due to aliasing, caused by the limited resolution of the
display screen. We will return to this problem in 4.3.2; we will discuss other particle rendering
methods, including techniques to add extra spatial information to particles in 4.3.5.
4.2.5 Environment Geometry
Display of environment geometry is often desirable in flow visualization, to show the context of
the particular flow problem studied, and also to support spatial orientation (figure 4.13). The
environment can usually be shown as a collection of surfaces or solid objects, allowing the use of
many spatial cues mentioned before. Techniques for rendering are mainly standard computer
graphics methods. But it is important to note that objects of the environment must be displayed
together with the data, and therefore the rendering technique must be used in combination with
other rendering techniques (see 4.3.4).

Figure 4.13 Environment geometry supports spatial orientation (Hin et al.’90)

4.2.6 Volume Rendering

If scalar data fields are associated with flow fields, or when scalar quantities are derived from a
flow field (see 3.3.6), volume rendering techniques can be used to visualize these data (for a
survey of volume rendering techniques, see Kaufman ‘90). In volume rendering, a scalar data
field (eg. a density field) is usually considered as a regular 3D array of cubical volume elements
(voxels), with one scalar data item associated with each volume element. A segmentation is made
of the data, by selecting only the voxels that satisfy a selection criterion (such as a given minimum
scalar value), or by applying other 3D image segmentation techniques. The selected voxels are
then projected onto the screen and rendered using special techniques for visible surface display and
shading.
Although in volume rendering surfaces are often displayed directly, iso-surfaces can also be
generated explicitly from volume data as a polygon mesh, using the method described in 3.3.8.

28
 If a surface is defined that intersects the data volume, the scalar values on the surface can be
determined by interpolation, and these can be shown using colour coding and iso-lines. The same
can be done with a surface at the model boundary, such as an airplane wing or a ship hull. The
surface may be shaded using directed light reflection, to provide good spatial cues.
Plane sections are often used in volume visualization to inspect a 3D data volume. This can be
achieved by moving a plane through the volume and watching the changing patterns projected onto
the section plane, to build a mental image of the whole field.

4 . 3 Special Rendering Techniques

4.3.1 Animation
Animation can be used for many purposes in flow visualization. As mentioned before, particle
motion can be very effectively visualized using animation sequences. The velocity of particles can
be directly visualized, provided that particle positions are known at constant intervals, and the
update time of the images on the screen is also constant.
Animations can either be produced in real time, or simply by playback of a series of pre-
computed images (frames). In real time animation, rendering of each frame is done during display.
This has the advantage that the animation may be interactively controlled by the user. A
disadvantage is that the screen update time usually depends on image complexity (for example, the
number of visible particles), which may vary per frame. Also, the choice of rendering techniques
is severely restricted. If we want to achieve the update rate of at least 5-10 frames per second
required to produce the illusion of motion, only the most simple and fast rendering techniques can
be used. To accelerate the process, some parameters such as particle paths can be pre-computed,
but this reduces the possibilities for interactive control.
Pre-computed frames may be displayed one at a time and recorded on a video tape, or played
back directly on the screen. In the latter case, all frames must be displayed from main memory, as
transfer from disk is too slow. The size of main memory thus restricts the length of an on-screen
animation. As the screen update rate is independent of the contents of the frames, playback speed
can be constant. Interactive control of most viewing parameters (such as viewing direction) or
positioning of particle sources is not possible.
A problem in animated flow visualization is the number of frames needed to get a clear view of
the flow patterns. For an unsteady flow, a new velocity field must be computed for each frame,
and new particle positions in this field. For one minute of animation, this must be performed for
about 1500 frames. With a steady flow, particle paths are derived from a single velocity field; as
the paths do not change in time, cyclicanimation can be applied (Van Wijk ‘90b; Stolk and Van
Wijk ‘91), showing a smooth motion by continuously repeated display of a limited number of
frames (typically 10-30).
Animation can be used for many other purposes in flow visualization. A changing viewpoint
(or a rotating object) provides a powerful depth cue (motion parallax), that is quite independent of
other cues, and can be used to resolve depth ambiguity even in wire frame drawings, such as
arrow plots. Other applications of animation are inspection by moving a section plane through an
area, or changing iso-values.
4.3.2 Aliasing and Anti-Aliasing
An important visual effect caused by the discrete nature of the display screen is known as aliasing.
It is obvious from the jagged edges and silhouettes, Moiré patterns, irregular and granular
textures, and missing small details. In animation, the effects are even more disturbing.

29
 Aliasing can be explained from signal theory (Blinn ‘89; Foley et al. ‘90). The colour of a
pixel is often determined by taking only one point sample. Better results can be achieved by
considering a pixel as a small area to which one colour is assigned, and determining this colour
from all objects that are visible within the pixel area. The contributions of the objects can be
estimated by taking several point samples for one pixel (supersampling), or by geometric
calculations (area sampling). The contributions are then combined to determine the colour of the
pixel.
When combining several samples to determine the colour for a given pixel, digital low-pass
filtering can be applied. The samples are weighted according to a filter function, which specifies
the weights dependent on the distance of a sample point to the pixel center. From the sample
values, a weighted average is then computed.
Analog to the spatial aliasing described above, temporal aliasing is caused by taking only one
point sample for each time step. This results in jerky motion, flicker, ‘strobing’ effects, and even
false movements that are unacceptable in animated flow visualization.
To reduce this, the time interval between two frames should be considered. For a moving
particle, this would be its trajectory over the time interval. Smooth motion can then be achieved
using digital filtering, using this trajectory or a number of samples on it. The result is a blurred
image of the moving particle, extended in the direction of motion, with a fuzzy shape. This effect,
known as motion blur, produces smooth motion without false effects (Sims ‘90).
4.3.3 Texture Synthesis and Texture Mapping
A texture is a pattern that can be mapped onto a surface, and used to modify the surface’s optical
properties. The most common form is modification of surface colour, resulting in a coloured
pattern pasted onto the surface; other properties, such as reflectivity, transparency, or normal
vector direction, can also be modified by texture (Heckbert ‘86). In texture mapping, obvious
adverse effects are caused by the aliasing problems mentioned above, and therefore digital filtering
is usually necessary to obtain good results (figure 4.14).

Figure 4.14 Aliasing in texture mapping - left: with point sampling; right: with digital
frequency filtering (Heckbert ‘86)

Existing images, such as digitized photographs, can be used as texture, but texture can also be
generated synthetically. Techniques for synthesis of texture has been mainly developed to achieve
naturalistic effects: to suggest certain materials or rough surfaces. As a primitive for visualization,
texture has been little explored, but some of its potential is shown by Van Wijk (‘90a, ‘91).
A good example is the synthesis of spot noise, a type of texture especially designed for data
visualization (Van Wijk ‘91). Spot noise is a stochastic texture generated as an addition of many
randomly distributed and weighted 2D patterns (spots). The texture can be globally and locally
controlled by varying the attributes of the spots. The spot size controls the granularity of the

30
texture: with small spots, it has the nature of white noise, while with large spots it has fractal
characteristics. This effect can be used to reflect a 2D data field, by scaling spot size according to a
scalar field. If the spots are scaled non-uniformly, the texture becomes non-isotropic, and thus
gets a directional nature. If the spots are locally scaled according to a 2D vector field, the texture
clearly shows the directional pattern. Other effects may be obtained by taking spots with different
shapes or patterns. In figure 4.15, two sets of lines and a velocity field are mapped onto the hull of
a ship.

Figure 4.15 Spot noise texture showing velocity on a ship hull (Van Wijk ‘91)

4.3.4 Hybrid Rendering

In scientific visualization it can be of particular importance to display several types of data in a
single picture. Data sets often contain combinations of scalar and vector data, and several types of
derived quantities may be computed from these data. Also, display of environment geometry is
necessary. Exploring the data may involve trying several combinations of visualization mappings,
which should not be restricted to rendering in separate modes. Quick and crude rendering should
be available as well as high-quality images. All these considerations lead to the concept of hybrid
rendering.
There are three approaches to achieve this. First, before rendering, all geometric primitives can
be converted to a single type. Examples of this are conversion of polygons and solids to
volumetric representations using 3D scan-conversion algorithms (Kaufman and Shimony ‘86), or
to convert volume data to polygons by surface reconstruction (Lorensen and Cline ‘87) or to
represent volume elements by transparent tetrahedra (Shirley and Tuchman ‘90). A second
approach is using a technique that can be adapted for rendering volumetric and surface data
directly, such as ray casting (Levoy ‘90). A third approach to hybrid rendering is rendering each
type of primitive with its own rendering technique, and then either merging the results using one
or more pixel buffers (Kaufman et al. ‘90, Ebert and Parent ‘90, Van Walsum et al. ‘91, Frühauf
‘91). Most of these techniques combine only volume and surface data (see eg. figure 4.13), and
do not consider particles as a separate type of primitive; an exception is Sims (‘90). Other types of
primitives may have to be added as well.
4.3.5 Advanced Particle Rendering
As observed in 4.2.4, particles may be considered as a special type of visual primitives, for which
special rendering techniques should be used. A first recognition of this idea was the development
of particle systems (Reeves ‘83, Reeves and Blau ‘85), where particles were used for modelling
and rendering ‘fuzzy objects’ with irregular, complex geometry, such as fire, trees and grass. An
early application of particle systems in 2D flow visualization was the animation of the atmosphere

31
of Jupiter in the film 2010 (Yaeger et al. ‘86). A more recent example was shown by Van Wijk
(‘90b).
A particle system is a collection of particles representing a fuzzy object. The particles have a
certain life cycle: they are born, have a limited lifetime, and die; during the lifetime, various
attributes such as motion dynamics (position, speed, motion direction), and visual appearance
(shape, size, colour, reflectivity, transparency) may vary as a function of time. They can also have
certain collective properties, which allows modelling of structured objects. In flow visualization,
the motion dynamics of the particles is determined by a flow field, but the other attributes can be
used to convey spatial information or to visualize associated data.
The rendering of the particles depends on their attributes. If particles are considered as light
emitting points, then rendering is simply done by adding intensities of particles projected onto a
single pixel, without any visibility calculations. These particles do not carry much spatial
information. If the particles are considered as light-reflecting, shading calculations must be
performed; also, self-shadowing (shadows cast by one particle onto others) is desirable as a depth
cue. Precise shading and shadowing calculations are prohibitive for very large numbers (more than
106), but then probabilistic shading models can be used. The position and orientation of a particle
determine the probability that it is lighted directly, and ambient, diffuse, and specular reflection
components are assigned, based on these probabilities (Reeves and Blau ‘85). The particles are
depth sorted for visibility, and rendered in back-to-front order. Transparency can thus be achieved
by blending the colours of the particles appearing in the same pixel.
To display more spatial information, a particle can be modelled as a very small surface element
that reflects directed light; it is then called a surface-particle (Stolk and Van Wijk ‘91). This has the
additional advantage that a (large) collection of surface-particles may show a stream surface or a
time surface (or indeed, any surface moving with and deformed by the flow), without
determination of explicit surface geometry. As an example, a stream surface consisting of surface-
particles is shown in figure 4.16.

Figure 4.16 Surface-particles (Stolk and Van Wijk ‘91)

32
 Shading calculations are performed for each particle, and for this purpose a normal vector is
attached to the particle. The direction of this vector is dependent on the desired type of surface.
The type of surface shown is determined by the shape of the particle source and the release time of
the particles. The particle source is a zero to three-dimensional geometric object, on which particle
starting positions are defined. The starting positions are distributed regularly or randomly over the
length of a line segment, the area of a polygon, or the surface of a solid. Particles are released
either continuously, or at discrete instants, at regular or random intervals. As space between the
particles is not filled, the surfaces are naturally transparent, depending on the density of the
particles.
A stream surface results if a line segment is used as a particle source, and particles are released
at random intervals. If a polygon, or a rectangle is used for the source, and particles are all
released at the same instant, a time surface is shown.
This technique is also very suitable to be used in animation. Particle release time and lifetime
are chosen such that no discontinuities show up in cyclic animation of steady flow. Animated
surface particles can give a very compelling view of 3D flow patterns.
Particle systems are often used in animations, and usually a kind of both spatial and temporal
anti-aliasing must be applied. The particles can be rendered as short line segments or small circles,
exploiting hardware facilities (Reeves and Blau ‘85). A more elaborate particle model has been
used by Sims (‘90). He used rounded, elongated shapes, with a distinct position and radius for
head and tail. The particles are scan converted, and spatial anti-aliasing and motion blur are
accomplished by decreasing opacity near the edges, and setting the head and tail positions in
accordance with velocity and shutter speed. Rendering of the particles is done without light
reflection.

5 Conclusions and Research Directions

In the preceding sections, we have reviewed different aspects of flow visualization, including
experimental and computer graphics flow visualization techniques. The connection between
experimental and computer-aided flow visualization is now beginning to develop. The current
strong demand for new flow visualization techniques, especially for large scale 3D numerical flow
simulations, can only be satisfied by combining the efforts of fluid dynamics specialists, numerical
analysts, and computer graphics experts. Additional knowledge will be required from perceptual
and cognitive psychology, and artists and designers can also contribute to this effort.
Flow visualization will not be restricted to techniques for giving an intuitively appealing,
general impression of flow patterns, but will increasingly focus on more specific physical flow
phenomena, such as turbulence, separations and reattachments, shock waves, or free liquid
surfaces. Also, purely visual analysis of flow patterns will be increasingly complemented by
algorithmic techniques to extract meaningful patterns and structures, that can be visualized
separately.
To take its place as a common research tool in fluid dynamics, computer graphics flow
visualization will also have to deal with more specific questions, such as validation of new flow
analysis techniques, involving comparison of data from CFD simulation results with experimental
data. This may be done visually by displaying both types of data with the same techniques, but
may also be supported by algorithmic similarity checking. Image processing techniques are often
applied to images acquired from experimental visualization (Yang ‘89) to derive data that can also
be computed in the CFD simulation or the visualization stage.

33
 Before they can be used as reliable research tools, the visualization techniques themselves
must also be carefully tested and validated. As we have seen in the previous sections, visualization
involves a sequence of many processing steps, where approximations are frequently used and
numerical errors can easily occur. Even at the visual level, interpolation artifacts, aliasing, or
contradictory depth cues may cause incorrect interpretations.
An important issue following development of visualization techniques, is the design and
implementation of flow visualization systems. Here, there are many other issues that should be
explored further, such as distributed processing, data management, standardization of data
formats, and user interaction. Many styles and modes of interaction are possible, depending on the
extent of control of the visualization (and flow simulation) process (Hearn and Baker ‘91). The
interaction style provided by general purpose visualization systems such as AVS (Upson et al ‘89)
and apE (Dyer ‘90), enabling the user to construct data flow networks, allows manipulation of the
visualization process rather than the displayable objects representing the data. The concept of
‘direct manipulation’ of the data needs clarification; desirable facilities would certainly include
interactive probing and interrogation (Speray and Kennon ‘90).
Research in computer graphics flow visualization is still in its early stages, and especially 3D
flow field visualization is still very much an open problem. At present, this is one of the great
challenges of scientific visualization. This calls for a cooperative effort in the development of new
techniques at all stages of the flow visualization process.

Acknowledgements
We would like to thank Delft Hydraulics, the Netherlands Energy Research Foundation (ECN)
and the High-Speed Laboratory of the Department of Aerospace Engineering at Delft University of
Technology for their cooperation and for permission to use several illustrations. Also, we want to
express thanks to Andrea Hin, Arthur Mynett, Guus Segal, and Jack van Wijk for their valuable
criticism of earlier versions of this paper. Finally, we thank the editor Heinrich Müller for his
encouragement and patience.

References
Batchelor, G.K. (1967) An Introduction to Fluid Dynamics, Cambridge University Press
Bertrand, F.H., P.A. Tanguy (1988) Graphical Representation of Two-dimensional Fluid Flow
by Stream Vectors, Communications in Applied Numerical Methods 4, pp. 213-217
Blinn, J.F.(1989) What We Need Around Here Is More Aliasing / Return of the Jaggy, IEEE
Computer Graphics and Applications 9(1), pp. 75-79, and 9(2), pp. 82-89
Briscolini, M., P. Santangelo (1991) Animation of Computer Simulations of Two-Dimensional
Turbulence and Three-Dimensional Flows, IBM Journal of Research and Development”
35(1/2), pp. 119-138
Bronsvoort, W.F., F.W. Jansen, F.H. Post (1991) Design and Display of Solid Models, In
Advances in Computer Graphics VI, G. Garcia, I. Herman (eds.), Springer Verlag, pp. 1-57
Bryson, S., C. Levit (1991) The Virtual Windtunnel: An Environment for the Exploration of
Three-Dimensional Unsteady Flows, In Proceedings Visualization ‘91, G.M. Nielson, L.
Rosenblum (eds.), IEEE Computer Society Press, pp. 17-24
Buning, P.G.(1989) Numerical Algorithms in CFD Post-Processing, von Karman Institute for
Fluid Dynamics, Lecture Series 1989-07
Chen, H.H., T.S. Huang (1988) A Survey of Construction and Manipulation of Octrees,
Computer Vision, Graphics and Image Processing 43, pp. 403-431
Chong, M.S., A.E. Perry, B.J. Cantwell (1990) A General Classification of Three-dimensional
Flow Fields, Physics of Fluids A 2(5), pp. 765-777
Dickinson, R.R. (1991) Interactive Analysis of the Topology of 4D Vector Fields, IBM Journal of
Research and Development 35(1/2), pp. 59-66
34
Dyer, D.S. (1990) A Dataflow Toolkit for Visualization, IEEE Computer Graphics and
Applications 10(4), pp. 60-69
Dyke, M. van (1982) An Album of Fluid Motion, The Parabolic Press
Ebert, D.S, R.E. Parent (1990) Rendering and Animation of Gaseous Phenomena by Combining
Fast Volume and Scanline A-buffer Techniques, Proceedings Siggraph ‘90, Computer
Graphics 24(4), pp. 357-366
Foley, J.D., A. van Dam, S. Feiner, J. Hughes (1990) Computer Graphics: Principles and
Practice, second edition, Addison-Wesley Publishing Company
Frühauf, M. (1991) Combining Volume Rendering with Line and Surface Rendering, In
Eurographics ‘91, F.H. Post, W. Barth (eds.), North Holland, pp. 21-32
Gallagher, R.S. (1991) Span Filtering: an Optimization Scheme for Volume Visualization of Large
Finite Element Models, In Proceedings Viualization ‘91, G.M. Nielson, L. Rosenblum (eds.),
IEEE Computer Society Press, pp. 68-75
Gallagher, R.S., J.C. Nagtegaal (1989) An Efficient 3-D Visualization Technique for Finite
Element Models and Other Coarse Volumes, Proceedings Siggraph ‘89, Computer Graphics
23(3), pp. 185-194
Geiben M., M. Rumpf (1991), Visualization of Finite Elements and Tools for Numerical
Analysis, In Advances in Scientific Visualization, F.H. Post, A.J.S. Hin (eds.), Springer
Verlag (to appear in 1992)
Gibson, J.J.(1950), The Perception of the Visual World, Houghton Mifflin Co.
Gibson, J.J.(1979), The Ecological Approach to Visual Perception, Houghton Mifflin Co.
Globus, A., C. Levit, T. Lasinski (1991) A Tool for Visualizing the Topology of Three-
Dimensional Vector Fields, In Proceedings of Viualization ‘91, G.M. Nielson, L. Rosenblum
(eds.), IEEE Computer Society Press, pp. 33-40
Haber, R.B., D.A. McNabb (1990), Visualization Idioms: A Conceptual Model for Scientific
Visualization Systems, In Visualization in Scientific Computing, G.M. Nielson, B. Shriver,
L.J. Rosenblum (eds.), IEEE Computer Society Press, pp. 74-92
Hearn, D.D., P. Baker (1991) Scientific Visualization, Eurographics ‘91 Tutorial Notes no. 6,
Eurographics Technical Report Series EG91 TN6.
Heckbert, P.S. (1986) Survey of Texture Mapping, IEEE Computer Graphics and Applications
6(11), pp.56-67
Helman, J., L. Hesselink (1989a) Automated Analysis of Fluid Flow Topology, 3D Visualization
and Display Technologies (Proc. SPIE) 1083, SPIE, Bellingham, Wash., pp. 825-835
Helman, J., L. Hesselink (1989b) Analysis and Visualization of Flow Topology in Numerical
Data Sets, IUTAM Symposium on “Topological Fluid Mechanics”, Cambridge, England
Helman, J., L. Hesselink (1989c) Representation and Display of Vector Field Topology in Fluid
Flow Data Sets, IEEE Computer 22(8), pp. 27-36
Helman, J., L. Hesselink (1990) Surface Representations of Two- and Three-Dimensional Fluid
Flow Topology, In Proceedings Visualization ‘90, A. Kaufman (ed.), IEEE Computer Society
Press, pp. 6-13
Helman, J., L. Hesselink (1991) Visualizing Vector Field Topology in Fluid Flows, IEEE
Computer Graphics and Applications 11(3), pp. 36-46
Hesselink, L., J. Helman (1987) Evalution of Flow Topology from Numerical Data, Invited
AIAA-paper, 87-1181-CP
Hesselink, L., J. Helman, K. Wu (1988) Visualization and Interpretation of 3-D Scientific Data
Sets, ICALEO ‘88-Conference, Santa Clara CA
Hibbard, W., D. Santek (1989) Visualizing Large Data Sets in the Earth Sciences, IEEE Computer
22(8), pp. 53-57
Hin, A.J.S., E. Boender, F.H. Post (1990) Visualization of 3D Scalar Fields using Ray Casting,
In Proceedings of the Eurographics Workshop on Visualization in Scientific Computing, M.
Grave, Y. Le Lous (eds.), to be published by Springer Verlag
Hultquist, J.P.M.(1990) Interactive Numeric Flow Visualization Using Stream Surfaces,
Computing Systems in Engineering, 1(2-4), pp. 349-353
Kaufman, A., ed. (1990) Volume Visualization, IEEE Computer Society Press
Kaufman, A., E. Shimony (1986) 3D Scan Conversion Algorithms for Voxel-Based Graphics, In
Proceedings 1986 Workshop on Interactive 3D Graphics, F. Crow, S.M. Pizer (eds.), ACM,
pp. 45-75

35
Kaufman, A., R. Yagel, D. Cohen (1990) Intermixing Surface and Volume Rendering, In 3D
Imaging in Medicine: Algorithms, Systems, Applications, K.H. Höhne, H. Fuchs, S.M. Pizer
(eds.), Springer Verlag, pp. 217-227
Kroos, K.A.(1984) Computer Graphics Flow Visualization Techniques for Three-Dimensional
Flow Visualization, In Frontiers in Computer Graphics, T.L. Kunii (ed.), Springer Verlag,
pp. 129-145
Levoy, M. (1988) Display of Surfaces from Volume Data, IEEE Computer Graphics and
Applications 8(3), pp. 29-37
Levoy, M. (1990) A Hybrid Ray Tracer for Rendering Polygons and Volume Data, IEEE
Computer Graphics and Applications 10(2), pp. 33-40
Lorensen, W.E., H.E. Cline (1987) Marching Cubes: a High Resolution 3D Surface Construction
Algorithm, Proceedings Siggraph ‘87, Computer Graphics 21(4), pp. 163-169
Merzkirch, W. (1987) Flow Visualisation, second edition, Academic Press Inc.
Nielson, G.M., B.S. Shriver, L.J. Rosenblum, eds. (1990) Visualization in Scientific
Computing, IEEE Computer Socitety Press
Papathomas, T.V., J.A. Schiavone, B. Julesz (1988), Applications of Computer Graphics to the
Visualization of Meteorological Data, Proceedings Siggraph ‘88, Computer Graphics 22(4),
pp. 327-335
Reeves, W.T.(1983), Particle Systems - a Technique for Modelling a Class of Fuzzy Objects,
ACM Transactions on Graphics 2(2), pp. 91-108
Reeves, W.T., R. Blau (1985) Approximate and Probabilistic Algorithms for Shading and
Rendering Structured Particle Systems, Proceedings Siggraph ‘85, Computer Graphics 19(3),
pp. 313-322
Sabin, M.A. (1985) Contouring - the State of the Art, in Fundamental Algorithms for Computer
Graphics, R.A. Earnshaw (ed.), Springer Verlag, pp. 411-482
Samet, H. (1990) The Design and Analysis of Spatial Data Structures and Applications of Spatial
Data Structures, Addison-Wesley Publishing Company
Schroeder, W.J., C.R. Volpe, W.E. Lorensen (1991) The Stream Polygon: a Technique for 3D
Vector Field Visualization, In Proceedings Visualization ‘91, G.M. Nielson, L. Rosenblum
(eds.), IEEE Computer Society Press, pp. 126-132
Sims, K.(1990) Particle Animation and Rendering Using Data Parallel Computation, Proceedings
Siggraph ‘90, Computer Graphics 24(4), pp. 405-413
Shirley, P., A. Tuchman (1990) A Polygonal Approximation to Direct Scalar Volume Rendering,
Proceedings San Diego Workshop on Volume Visualization, Computer Graphics 24(5), pp.
63-69
Speray, D., S. Kennon (1990) Volume Probes: Interactive Data Exploration on Arbitrary Grids,
Proceedings San Diego Workshop on Volume Visualization, Computer Graphics 24(5), pp.
5-12
Stolk, J., J.J. van Wijk (1991) Surface-particles for 3D Flow Visualization, In Advances in
Scientific Visualization, F.H. Post, A.J.S. Hin (eds.), Springer Verlag (to appear in 1992)
Strid, T., A. Rizzi, J. Oppelstrup (1989), Development and Use of some Flow Visualization
Algorithms, von Karman Institute for Fluid Dynamics, Lecture Series 1989-07
Thorpe, S.J. (1990), Image Processing by the Human Visual System, Eurographics ‘90 Tutorial
Notes no. 4, Eurographics Technical Report Series EG90 TN4
Upson, C. et al. (1989), The Application Visualization System: a Computational Environment for
Scientific Visualization, IEEE Computer Graphics and Applications 9(4), pp. 30-42
Upson, C., M. Keeler (1988) V-Buffer: Visible Volume Rendering, Computer Graphics 22(4),
pp. 59-64
Walsum, T. van, A.J.S. Hin, J. Versloot, F.H. Post (1991) Efficient Hybrid Rendering of
Volume Data and Polygons, In Advances in Scientific Visualization, F.H.Post, A.J.S. Hin
(eds.), Springer Verlag (to appear in 1992)
Wavefront Technologies (1990) The Data Visualizer Version 1.0 Users Guide, 1990
Wijk, J.J. van (1990a) Rendering Lines on Curved Surfaces, In Proceedings of the Eurographics
Workshop on Visualization in Scientific Computing, M. Grave, Y. Le Lous (eds.), to be
published by Springer Verlag
Wijk, J.J. van (1990b) A Raster Graphics Approach to Flow Visualization, In Eurographics ‘90 ,
C.E. Vandoni, D.A. Duce (eds.), pp. 251-259
Wijk, J.J. van (1991) Spot Noise - Texture Synthesis for Data Visualization, Proceedings
Siggraph ‘91, Computer Graphics 25(4), pp. 309-318

36
 Wijk, J.J. van (1992) Rendering Surface Particles, paper submitted for publication,
Yaeger, L., C. Upson, R. Myers (1986) Combining Physical and Visual Simulation - Creation of
the Planet Jupiter for the Film "2010", Proceedings Siggraph ‘86, Computer Graphics 20(4),
pp. 85-93
Yang, W.J., ed. (1989) Handbook of Flow Visualization, Hemisphere Publishing Corporation

37

View publication stats