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

net/publication/287218131

Fractal Objects in Computer Graphics

Conference Paper · June 2015

CITATIONS READS

2 8,640

3 authors, including:

Costin-Anton Boiangiu
Polytechnic University of Bucharest
169 PUBLICATIONS   590 CITATIONS   

SEE PROFILE

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

Software Projects Management View project

Computational Geometry View project

All content following this page was uploaded by Costin-Anton Boiangiu on 18 December 2015.

The user has requested enhancement of the downloaded file.

 Fractal Objects in Computer Graphics
COSTIN-ANTON BOIANGIU, ADRIAN GABRIEL MOROSAN, MARIAN STAN
Computer Science Department
“Politehnica” University of Bucharest
Splaiul Independentei 313, Sector 6, Bucharest, 060042
ROMANIA
[email protected], [email protected], [email protected]

Abstract: This paper presents methods that can be used in generating an entire planet from mathematical
objects, possibly starting from a small random seed. The planet will be generated only from specified
mathematical objects, fractals, procedural models and constructive solid geometry. This planet will include
elements such as vegetation, mountains, water, waves, rocky and sandy soil and clouds.

Key-Words: fractal geometry, fractal generation procedures, water generation, fractal vegetation, random
fractal terrain, fractal-generating software, computer graphics.

1 Introduction symmetric.
Each bump is certainly longer than the initial
segment, but still contains finite space inside. What
1.1 History is strange is that the perimeter moves towards
Unconventional mathematician Benoit Mandelbrot infinity rather than to converge to a special number.
created in 1975 the term fractal from the Latin word Mandelbrot saw this and used this example to
"fractus", which means irregular or fragmented. explore the concept of fractal dimension and to
These irregular and fragmented shapes are all prove that the measuring of the coast is an exercise
around us. Fractals are a visual expression of a of approximation. [3]
repeating pattern or formula that starts simple and
becomes progressively more complex. [1]
One of the first applications of fractals emerged 1.2 Characteristics
long before the term had been created. Lewis Fry The characteristics of fractals were for the first time
Richardson was an early twentieth century English indicated by Alain Boutot: “It has a fine structure
mathematician who studied the length of the coast with details on all scales of observation; It is too
of England. He reasoned that the length of the irregular to be described in the language of
coastline depends on the length of the measuring Euclidean geometry, both locally and globally; It is
instrument. Measuring with a measuring instrument a self-similar structure: is the analogon of the whole;
gives a length, but considering the more irregular It has fractal dimension higher than the topological
coastline by measuring with a smaller apparatus dimension”.
gives a longer length. [2]
If there is a philosophical conclusion to reach,
1.2.1 Self-similarity
there is an infinite coastline containing a finite
space. The same paradox was presented by Helge
von Koch’s snowflake. [3]

Fig. 1 Helge von Koch’s snowflake (2008) [3]

This fractal involves taking a triangle and

transforming each central third of each segment in a Fig. 2 Sierpinski Gasket [16]
triangular bump, in a way that makes the fractal
Property of self-similarity means that the parts are 1.3 The elements of a fractal
similar to the whole, with variations. Fig.2 contains, A vector-base fractal is composed of two parts: the
on the left side, a zoomed-out image of Sierpinski initiator and the generator. In Fig.3, are represented
Gasket, the poster child of fractals, and on the right the generator and initiator for Koch Snowflake. It
side, there is a zoomed-in image, revealing the starts with an equilateral triangle as the initiator and
scaling and self-similarity the characterize a line that is divided into three equal segments as the
fractals.[17] generator.

1.2.2 Scaling
Because of self-similarity, features at one spatial
resolution are related at other spatial resolutions.
The smaller features are smaller copies of the larger
features. The length at finer resolution will be
longer because these finer features are included.
How the measured properties depend on the Fig. 3 Generator and Initiator (image taken from [19])
resolution used to make the measurement is called
the scaling relationship. [17] The first iteration is realized by replacing every
line of the initiator with the full generator. A
snowflake can be approximated by iterating this
1.2.3 Bounded infinity operation again and again (Fig.4), replacing every
Bounded infinity means that one can trace infinite line of the new initiator with the full generator. To
length within a finite boundary, as demonstrated in generate a real Koch Snowflake is necessary an
the Koch snowflake’s infinite line length infinite process, but in practice, the process ends
circumscribing a finite area. after a finite number of iterations. [19]
In other words, the fractal object can be
represented by using a recursive function:

x, f x , f  f x ,... (1)

The recursive process ensures that connections

will be made, but we see only parts of the whole. Fig. 4 The Koch Snowflake – Iterations (image taken
[17] from [19])

1.2.3 Fractal dimension 1.4 The generation of fractal objects

In Euclidian geometry, one dimension is represented For generating fractal objects are several techniques,
by a line. In two dimensions, it is Cartesian place all these techniques are using feedback processes
and in higher dimensions, it is a coordinate space (Fig.5) in which the output of one iteration is used
with three or more integer number coordinates. as input for the next one. [20]
Mandelbrot said that the fractals have the property
of fractional dimensions. For a self-similar fractal,
the dimension (Hausdorff dimension, named after
mathematician bearing the same name) is defined
as:

s d  c (2)
Fig. 5 Diagram of a feedback process by N. Mukaia et
, where d is the Hausdorff dimension, c the al. [20]
amount of new copies you get after one iteration and
s is the scaling factor. Using the logarithmic laws,
the formula is: [18] 1.4.1 Generation of fractals using formal
languages
log c These languages, which are used for
d (3)
biomorphological descriptions, are known as L-
log s
systems (Lindenmayer systems). Filamentous
organisms can be formalized and described by From (6) and (7), using decomposition the complex
combining L-systems with branching patterns. mapping into real and imaginary part:
Firstly, L-systems generate strings in a feedback
loop (Fig.5). Then, in order to translate the strings  xn 1  f  xn , yn 
 (8)
 yn 1  g  xn , yn 
into morphological description, additional drawing
rules are necessary. This translation is necessary
because the strings do not contain geometric
information. [20] For each point in the complex field it can be
Formal languages are the most suitable for determined how many iterations are needed until a
generating plant-like objects. point escapes to other points of attraction or towards
infinity. For both cases are formulated stopping
criteria and for each mapping, it is possible to count
1.4.2 Generation of fractal objects using Function how many iterations are necessary until one or more
Systems of these criteria are reached. The number iterations
A second approach for generating fractal objects is a is used as an argument for color function. [20]
method based on Iterated Function Systems (IFS).
This method is used for a large class of fractal
objects and it may also be applied for modelling 2 Algorithms
natural objects. The IFS consists of a d-dimensional In this paper will be presented ways for building
space set of mapping into itself (4) and a set of objects like water waves, lands with mountains,
corresponding probabilities (5). [20] clouds, vegetation, soil with sand or rocks and land
textures, generated mathematically. [4]
M  M 1 , M 2 , M 3 ,..., M n  (4)

n 2.1 Generation of water

P  P1 , P2 , P3 ,..., Pn , in which:  P 1 i (5) To generate water waves there are several
i 1 algorithms. The main types of waves used are sine
waves, Gerstner waves and FFT waves. There are
also algorithms for generating turbulent waves,
1.4.3 Geometric construction of fractals algorithms for generating waves for shallow water
Many fractal objects may be generated with and algorithms for generating movement of
geometric constructions. The class of fractals incompressible viscous fluids using the Navier
created using this method is known as linear fractals Stokes equations.
class. A part of linear fractals can be represented by Everything can be calculated mathematically,
the initiator (initial polygon), generator and the including water color and its reflection and
production rule. In addition to the production rules refraction. There are also algorithms for generating
geometric information is also necessary, which is caustic and foam, water drops and bubbles, using
stored in data structures. systems of particles. [4]
The main advantage of this class of fractals is
that geometric production rules can be designed
interactively and in a, relatively, easy way. [20] 2.2.1 Sine Waves
Sine waves are calculated using sums of sines,
which are continuous functions that describes the
1.4.4 Non-linear complex mappings generation height and orientation of the water surface at all
The most important part of the fractals is points of the plane.
represented by fractals in which relation between The status of each wave according to the
input and output (Fig.5) is non-linear, using horizontal position  x, y  and time t is defined as:
complex variables and parameters. Two examples of
Wi x, y, t   Ai  sin Di  x, y   wi  t  i 
fractals included in this class are: Mandelbrot and
Julia sets. [20] (9)

zn 1  F zn  (6) Where:

Ai is the amplitude,
F  f  ig (7) Di is the direction, the horizontal vector that is
perpendicular to the wave front which moves along
the top of the wave, waves are similar to sine waves.
wi is the frequency,
 i is the phase constant and
  S   , where S is the speed, also known as
the distance the wave travels at the top of a frame,
and   2 / L , where L is the length of the wave,
that is the distance between the peaks of the waves.
For all  x, y  in the horizontal plane 2D, the 3D
position of the sine waves surface is:

Px, y, t   x, y, H x, y, t  (10)

where H x, y, t  is the total area, and is defined

Fig. 6 Gerstner Waves (2007) [4]
as: H x, y, t   Wi x, y, t  . [4]
Unlike sine waves, the Gerstner waves and
points move sideways and even if the overall wave
2.1.2 Gerstner waves height is the same, Gerstner waves are more realistic
Unlike sine waves, which have a rounded because sharp waves can be obtained. [4]
appearance and are suitable for calmer waters,
Gerstner waves can control steepness, thus sharp
waves can be generated, making them more suitable 2.1.3 FFT waves
for rough waters. FFT waves are not based on any physical model, but
For all  x, y  in the horizontal plane 2D, the 3D
they are based on statistical models on actual
observations of oceans and seas. These waves have
position of the Gerstner surface waves is: been used commercially several times, especially
for movie animation seas and oceans.
  H '  x, y, t   H '  x, y, t  
P x, y, t    x  ,y , H  x, y, t  In statistical models and crafts, wave height is a
 x y  random variable, being a function of horizontal
(11) position and time. In order to obtain FFT waves the
wave height field is split into a set of sine waves
where t is the time, with different amplitudes and phases, after which
 H '  x, y, t  inverse FFT transform is used to rapidly assess the
is the partial derivative of amount obtained. [4]
x
H ' x, y, t  in x direction and is defined as:
H ' x, y, t  2.1.4 Agitated waves
  Qi Ai  Di .x  coswi Di  x, y   i t  The FFT algorithm for generating waves produces
x
waves that have rounded edges, suggesting calmer
,
 H '  x, y, t 
weather. To get restless waves, with sharper peaks
is the partial derivative of and flatted bases, instead of directly modify the
y height field, positions of the points need to be
H ' x, y, t  in y direction and is defined as: moved horizontally.
H ' x, y, t 
FFT waves are an alternative for sine waves,
  Qi Ai  Di . y  coswi Di  x, y   i t  while the restless waves are an alternative Gerstner
y waves, and although the algorithms to generate the
, waves are different, the results are similar. [4]
H ' x, y, t   H x, y, t   Qi / wi ,
and H  x, y, t  H (x, y, t), Wi , Di , Ai and  i 2.1.5 Waves for shallow waters
are the same as in the case of sine waves. For shallow water, waves are generated with the
In addition to the sine waves, Gerstner waves Saint Venant equations that are partial hyperbolic
control the inclination of the wave through the Qi differential equations which describe the flow of a
parameter. If its value is zero the obtained Gerstner fluid underneath a surface of pressure. [4]
2.1.6 Fluid Movement 2.1.10 Foam, drops and water bubbles
Incompressible viscous fluid motion is described When the water surface is very agitated or when it
entirely by the Navier Stokes equations. In these encounters obstacles it generates foam, water drops
equations there are three types of forces that act: and bubbles due to breaking waves. This can be
body forces, pressure and viscous forces. achieved by means of a system of particles that will
Body forces are the forces acting on the entire be based on Newtonian dynamics. [6]
surface of the water. It is generally assumed that
these forces are formed only from gravity. Pressure
forces act to the inside of the fluid and to the surface 2.1.11 Generation of rivers
normal, and the forces due to friction of the water The classical method for generating rivers is a part
are the viscous forces and they act in all directions of height-map generation algorithm. The algorithm
over the entire surface of the water. [5] generates a terrain model around a precomputed set
of ridge lines and rivers network. First, it is used a
rapid method that generates the ridges and rivers
2.1.7 Water Color network. Then, an extension of the basic midpoint
The color of water is given by the reflection and displacement method is applied for generating
refraction general, but the water may also have a fractal terrain model around a pre-filled ridges and
color which depends on the direction of the incident rivers network. [15]
light beam, the direction of viewing and the
properties of the water. [5]

2.1.8 Reflection and refraction of water

Most visual effects of the water are caused by the
reflection and refraction. This occurs when a ray
strikes the surface of the water, and part of it is
reflected back into the atmosphere and the other part
is refracted in the volume of water.
Fig. 8 The Midpoint Displacement's Inverse process by
One of the most important visual aspects of
F. Belhadj et al. (2005) [15]
realistic rendering of water through the Fresnel
equation which defines a factor of combination In Fig.8, the left image shows: in black color the
between reflection and refraction. [5] computed elevations and in the white color the
elevations that still have to be computed. The right
image, in black/gray color is represented the
tridimensional view of D.E.M, in red the elevations
computed with the M.D.I. process and in white the
elevations that still have to be computed. [15]
The second approach is a post-processing step on
the existing height-map. For the first one, a
generated river network forms a basis from which a
height-map is inferred. For the last one, a height-
map is analyzed to find the potential stream routes
from mountains into valleys. [12]
Fig. 7 Ocean with Gerstner waves, reflection and
refraction by Jerry Tessendorf (2005) [5]
2.1.12 Generation of oceans and lakes
Fig.7 presents Gerstner wave generated water, Procedural water bodies, such as oceans and lakes
and water color formed by reflection and refraction. and theirs connections, stream networks, deltas and
waterfalls are similarly generated. Oceans are
commonly generated setting a fixed water level and
2.1.9 Caustics for lakes, the classical method is a flooding
Caustics result from reflected or refracted light rays algorithm from points of low elevation. [12]
on a curved surface and therefore they focus only in
certain areas of the receiving surface. They can be
generated “fractally” or mathematically. [6]
2.2 Generation of land
For the generation of fields with mountains there are
multiple algorithms. [7]

3.2.1 Transcendental land

Transcendental lands are lands mathematically
calculated using sine and cosine functions.
Transcendental lands have a rounded appearance
and relate to sine waves resembling water. [7]

3.2.2 Fractal land

Fractal terrains are more realistic than the
transcendental terrains, due to the sharp mountain
peaks. These fields can be generated using several
algorithms, among which the most important are the
median shift algorithm, Diamond-Square algorithm Fig. 9 Fractal terrain generation using Diamond-Square
and FFT. [7] algorithm by David P. Feldman (2012) [8]

2.2.3 Median shift algorithm 2.2.5 FFT algorithm

Median shift algorithm provides realistic results. Unlike the movement of the median algorithm and
This algorithm assumes that the start is a line Diamond-Square algorithm which have linear
between two points, then the median of that line is running time, the FFT algorithm’s runtime is
moved with a random value in the vertical direction, logarithmic.
then the medians of the two new segments are Another difference of the FFT algorithm is that
moved with random values in the vertical direction the terrain has a more rounded texture and has no
and then the previous step is repeated until the raised or pointed peaks. [9]
desired level of detail is reached. [8] At distance either less frequencies can be
considered, or the terrain simplified for speed
purposes. [11]
2.2.4 Diamond-Square algorithm
The Diamond-Square algorithm generates more
realistic terrain than the previous algorithm, because 2.2.6 Erosion fractal
the latter one leaves square objects in the field. The The height-maps can be transformed using
Diamond-Square algorithm mitigates this by simulations of physical phenomena, such as erosion.
alternating calculated values of the middle points of In order to diminish sharp changes in elevation, it is
squares and diamonds. used thermal erosion, by iteratively distributing
This algorithm assumes the start by assigning a material from higher to lower points, until the
random heights of the four corners of the grid, and maximum angle of stability for a material is
then averaging four corners, plus a random reached.
disturbance and assigns this value to the middle Another type of erosion is caused by rainfall or
point of the square formed by four points. Then it fluvial erosion. This type can be simulated using,
takes each diamond achieved, calculate the average cellular automaton (model of a system of “cell”
of the four corners of each diamond, plus a random objects), where dissolved material that flows out to
disturbance and assigns the middle point of the other cells and the amount of water are calculate
diamond. The previous two steps are then repeated based on the local slope of the elevation profile. [12]
until reaching the desired level of detail. [8]
Fig.9 shows the generation of a fractal terrain
using Diamond-Square algorithm, from the initial 2.2.6 Digital elevation models (DEM) method
stages of low level detail in the final stages with Besides other methods, this setup allows the user to
high level of detail. [8] interactively edit the height-map. A user draws a 2D
map of polygonal regions, each of which is marked
to have a certain elevation profile. The straight
boundaries of the regions are perturbed and
rasterized in a grid. Then, for each region, DEM flowers, fruits, vegetables or leaves. Plants can be
data is selected to match the requested elevation generated using L-systems or using iterated function
profile. The selection of data is realized using systems. [4]
genetic algorithms. Branching patterns may be defined as L-systems
The main advantages of this method are: realism, by using K denoted by <G, W, P>, in which G is a
extensibility and ease of use (intuitive control, with set of symbols, W is the starting string and P is the
low input requirements). Although, the generated production rule. Monopodial branching, in which
transitions at the boundaries between regions are the growth of the main axis continues throughout
still rather abrupt. [13] the plant’s life, may be represented by the following
L-system [20]:

2.1 Cloud Generation KM  G,WM , PM (12)

To generate clouds, noise functions are used in
G  0,1, [, ]
general. The noise that offers the most realistic
results is Perlin noise. Meteorological phenomena, (13)
such as lightning, can be mathematically generated
or through fractals. Rain and snow, can be generated WM  0 (14)
using a particle system. [4]
PM  0  1[0]0,1  1, [ [, ] ] (15)

2.3.1 Noise functions

The role of noise functions is to provide a pseudo- <Iteration> <Generated String>
random signal, efficiently implemented and
repeatable over a three-dimensional space. In
general, the sound functions receive an integer as a 1: 0
seed and they return a pseudo random number based
on the received parameter. [4]
2: 1[0]0

2.3.2 Perlin Noise

3: 1[1[0]0]1[0]0
To create a Perlin noise function, a noise and an
interpolation function are required. The noise
function must be, in general, a random number 4: 1[1[1[0]1]1[0]0]1[1[0]0]1[0]0
generator that will return the same value if two
numbers with the same value are sent as input and The drawing rule showed in Fig.10 is one
different values if two different values are sent as a possible way to visualize the generated string.
parameter.
A standard interpolation function receives three
parameters, two of which represent the values
between the return value that must be interpolated
and the third parameter based on which returned
value is interpolated. [10]

2.4 Generation of vegetation

Vegetation can also be generated with fractals or
mathematically. Procedural vegetation is a classic
research topic in the field of procedural modelling Fig. 10 Drawing rule for the string 1[0]0, shape A and B
by N. Mukaia et al. [20]
and includes both procedures for generating 3D
plant models and trees and specific methods for
An alternative system to procedurally model
placement on a given surface. [4]
plants is by placing plant components in a graph.
Connected components can be structured in sub-
2.4.1 Generating plants graphs. The system traverses this graph, generating
There are many plants that can be generated by and placing instances of the components in an
fractals or mathematically, including ferns, grass, intermediate graph that is used for geometry
generation. A set of components is connected by the placing some fundamental patterns at random. The
user to describe the structure of the plant. The more patterns are used, the more varieties are
algorithms are controlled by graphical user interface generated. On the other hand, the weathered image
on the basis of spline functions. [14] is generated using fractal functions. Intended
images, such as a texture that represents water drop
track on the wall, can be realized by selecting the
2.4.2 Generation of trees fractal seeds. Also, combining both material and
Trees can be generated using L-systems or iterated weathered images can make more realistic images
function systems. L-system is a classical and often for texture mapping. The generated images look like
used example of a rewriting system. Although the very natural and enable the modelling data of
L-systems are used for rewriting strings of text, the landscape simulation very realistic. [20]
resulting set of symbols can be interpreted in 2D
and 3D. One of the best known such tree is the tree
of Pythagoras, which is built recursively. [4] 2.8 Generation of other elements
The most important of these elements are animals
such as spiders, snails, peacocks, sea urchins and
2.5 Soil generation starfish, but the generation of stalagmites and
The most important elements of this category that stalactites and crystals is also possible. [4]
can be generated are rocks and sand. Sand dunes
using noise functions can also be generated. [4]
3 Conclusions
This paper presents ways to generate as many
2.6 Urban environments classes of objects on the basis of equations, possibly
The common approach for procedurally generating
with minimal data, such as a seed. Items that can be
cities is to start from a dense road network and
generated include: terrain, terrain textures, clouds,
identify the polygonal regions enclosed by streets.
vegetation, soil and water.
Building lots are the result of subdivision of these
regions. Then, for populating these lots, the lot
shape is used directly as the footprint of a building Acknowledgments:
(Fig.11). Another method is to fit the building The research carried by the first author was
footprint on the lot. By simply extruding the supported by Electronic Arts Romania S.R.L.
footprint to a random height, it can be generated a
city of skyscrapers and office buildings. These
approaches are used to create a macro environment, References:
for more complex details, are necessary several rule- [1] Benoit Mandelbrot, The Fractalist: Memoir of a
based methods. [12] Scientific Maverick, Knopf Doubleday Publishing
Group, 2012
[2] Lewis Fry Richardson, Oliver M. Ashford, Philip
Gerald Drazin, The Collected Papers of Lewis Fry
Richardson, Cambridge University Press, 2009
[3] NOVA, Hunting the Hidden Dimension, PBS, 2008,
http:// www.pbs.org/ wgbh/ nova/ physics/ hunting-
hidden-dimension.html, accessed 26.05.2015
[4] NVIDIA, GPU Gems, NVIDIA Corporation, 2007,
Fig. 11 Procedural sketching session: a) river flowing https:// developer.nvidia.com/ gpugems/ GPUGems/
towards the sea, b) road feature crossing the river, c) city , accessed on 26.05.2015
created along the river banks by R.M. Smelik (2011) [12] [5] Jerry Tessendorf, Simulating Ocean Water,
Computer Graphics Laboratory, 2005
[6] Yaohua Hu, Luiz Velho, Xin Tong, Baining Guo,
2.7 Generation of textures Harry Shum, Realistic, Real–Time Rendering of
They can be generated using height maps obtained Ocean Waves, Computer Animation And Virtual
at the terrain generation. [4] Another method is used Worlds, vol. 17, issue 1, pp. 59-67, DOI: 10.1002
for generating texture image such as surfaces of [7] Krista Bird, Thomas Dickerson, Jessica George,
houses or roads. For this, random or fractal Techniques for Fractal Terrain Generation,
functions are used. This method uses two types of HRUMC, 2013
image. One is a material image and the other is a [8] David P. Feldman, Chaos and Fractals: An
weathered one. The material image is generated by Elementary Introduction, Oxford University Press,
 2012 landscapes”
[9] Michael F. Worboys, Matt Duckham, GIS: A [15] F. Belhadj, P. Audibert. Modeling Landscapes with
Computing Perspective, Second Edition, CRC Press, Ridges and Rivers, November 2005, Proceedings of
2004 the 3rd international conference on Computer
[10] David S. Ebert, Texturing & Modeling: A graphics and interactive techniques, Australasia and
Procedural Approach, ISBN-13: 978-0122287602, South East Asia, pp. 447-450
1994 [16] https://classes.yale.edu/fractals/IntroToFrac/SelfSim
[11] Costin-Anton Boiangiu, Bogdan Raducanu. “3D /SelfSim.html accessed 27.05.2015
Mesh Simplification Techniques for Image-Page [17] L. Strudwick. Infinite Space and Self-Similar Form
Clusters Detection”. WSEAS Transactions on in Alchemy and Fractal Geometry, Mythological
Information Science, Applications, Issue 7, Volume Studies Journal, Vol 1, No 1 (2010)
5, pp. 1200 – 1209, July 2008 [18] Magdy M. Ibrahim and Robert J. Krawczyk.
[12] R.M. Smelik. “A Declarative Approach to Generating Fractals Based on Spatial Organizations.
Procedural Generation of Virtual Worlds” , 30 College of Architecture, Chicago, IL USA
november 2011, thesis [19] M.M. de Ruiter (Ed.). “Advance in Computer
[13] Ryan L. Saunders. “Terrainosaurus: Realistic Graphics III”. EurographicSeminars
Terrain Synthesis Using Genetic Algorithms”, [20] N. Mukaia , Y. Sakaguchia, M. Kosugia. “A Method
December 2006 for Generating Texture Images used on Landscape
[14] O. Deussen, C. Colditz , L. Cocunu , H.C. Hege. Simulation”. Electronic & Computer Engineering,
“Efficient modelling and rendering of synthetic Musashi Institute of Technology, Tokyo, Japan

View publication stats