International Journal of Pattern Recognition and Artificial Intelligence Vol. 15, No. 7 (2001) 1007–1021 c World Scientific Publishing Company GENERATING ISOTROPIC DISCRETE WAVES ON CELLULAR AUTOMATA FABIEN FESCHET LASS Laboratory, UMR 5823 – MA2D – Universit´ e Lyon 1, Bat 101, 43 Bd du 11 Novembre 1918 – 69622 Villeurbanne Cedex, France [email protected]LAURE TOUGNE E.R.I.C. Laboratory, Bat L – Universit´ e Lyon 2, 5 av. Pierre Mend` es-France, 69676 Bron Cedex, France [email protected]Cellular automata are a massively parallel computation model with discrete time and local rules. They are well adapted to biological or physical simulations. However, they are intrinsically anisotropic. The possibility of computing isotropic figures on cellular automata such as circles has already been proved. 4 Moreover, the previous construction enables to compute all the major discretizations known in the literature. We present in this article an extension of this work to the construction of spheres in three dimensions. A local characterization of a sphere is presented based upon the relationship between spheres and circles. This leads to the possibility of constructing a family of concen- tric discrete spheres in real time. Moreover, the approach can use many discretization schemes leading to the construction of various discrete spheres as done for circles. Keywords : Discrete sphere; 3D cellular automata; discrete circle; discrete waves. 1. Introduction A cellular automaton is a regular homogeneous network of identical simple machines which locally communicate with each other. The subjacent network is the graph Z n so that cellular automata are often viewed as a discrete massively parallel com- putation model. All machines evolve from one state to another at the same top following a discrete time. The dynamic of cellular automata is very sensitive to the initial conditions. This last property leads physicians or biologists to use cellular automata to model real phenomena with the goal of predicting evolution. However, the physical universe is usually seen as a three-dimensional continuous and isotropic space through the classical Euclidean metric. By using cellular automata, a discrete universe is considered. This is able to have several behaviors but the isotropy is not natural since directions are privileged. Many physical phenomena are anisotropic, but some of them, such as waves propagation, are isotropic which means that their space extension is not direction-depending. 1007 Int. J. Patt. Recogn. Artif. Intell. 2001.15:1007-1021. Downloaded from www.worldscientific.com by 159.84.103.173 on 03/14/13. For personal use only.
15
Embed
GENERATING ISOTROPIC DISCRETE WAVES ON CELLULAR AUTOMATA · 2. Cellular Automata Let us now remember some classical de nitions concerning the cellular automata. We rst recall the
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Cellular automata are a massively parallel computation model with discrete time andlocal rules. They are well adapted to biological or physical simulations. However, theyare intrinsically anisotropic. The possibility of computing isotropic figures on cellularautomata such as circles has already been proved.4 Moreover, the previous constructionenables to compute all the major discretizations known in the literature. We present inthis article an extension of this work to the construction of spheres in three dimensions.A local characterization of a sphere is presented based upon the relationship betweenspheres and circles. This leads to the possibility of constructing a family of concen-tric discrete spheres in real time. Moreover, the approach can use many discretizationschemes leading to the construction of various discrete spheres as done for circles.
Keywords: Discrete sphere; 3D cellular automata; discrete circle; discrete waves.
1. Introduction
A cellular automaton is a regular homogeneous network of identical simple machines
which locally communicate with each other. The subjacent network is the graph
Zn so that cellular automata are often viewed as a discrete massively parallel com-
putation model. All machines evolve from one state to another at the same top
following a discrete time. The dynamic of cellular automata is very sensitive to the
initial conditions. This last property leads physicians or biologists to use cellular
automata to model real phenomena with the goal of predicting evolution. However,
the physical universe is usually seen as a three-dimensional continuous and isotropic
space through the classical Euclidean metric. By using cellular automata, a discrete
universe is considered. This is able to have several behaviors but the isotropy is not
natural since directions are privileged. Many physical phenomena are anisotropic,
but some of them, such as waves propagation, are isotropic which means that their
space extension is not direction-depending.
1007
Int.
J. P
att.
Rec
ogn.
Art
if. I
ntel
l. 20
01.1
5:10
07-1
021.
Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
159
.84.
103.
173
on 0
3/14
/13.
For
per
sona
l use
onl
y.
October 25, 2001 9:23 WSPC/115-IJPRAI 00130
1008 F. Feschet & L. Tougne
By constructing isotropic figures with a cellular automaton, we refer to the
construction of an approximation of an isotropic object. In particular in Refs. 4
and 7, it has been proved that we can construct discrete circles in real time by
cellular automata. Starting from an initial configuration in which all cells of the
discrete plane are in a same state except one (the center of the concentric circles),
the cellular automaton highlights at time t the cells that belong to a digitization
of the circle of radius t. In fact, it is possible to construct the families of all the
well-known good digitizations of circles found in the literature.
The goal of this article is to extend the previous works and more precisely to
present how we can generate discrete isotropic waves in dimension two (circles) and
in dimension three (spheres) with the help of cellular automata. In fact, whatever
the dimension is, the principle remains the same: we have to find a local charac-
terization of the cells that belong to a wave at a given time. Then each cell, in the
neighborhood of a cell belonging to the discrete wave at time r, enters the state
wave at time r+ 1 according to the states of all its neighbors at time r and the set
of transition rules.
The following section deals with definitions concerning cellular automata. In
Sec. 3, we explain the strategy used for the construction. This leads us to the
construction of figures, called paraboloids, in Sec. 4, from which we can construct
waves in Sec. 5. Finally we conclude and present some future works.
2. Cellular Automata
Let us now remember some classical definitions concerning the cellular automata.
We first recall the definition itself of a cellular automaton and the notion of con-
figuration. Then we introduce the notion of signal used to describe the behavior of
a cellular automaton. Finally, we define the notions of construction of figures and
construction in real time.
2.1. Standard definitions
Definition 1. An n-dimensional cellular automaton (or n-CA), A, is a 4-uplet
(n, S,B, δ) such that:
• n is the dimension of the subjacent grid,
• S is a set of elements of which are the states of A,
• B is the neighborhood of A and is a finite subset of Zn, with cardinal |B|,• δ is a function from S|B| to S, called the local transition function of A.
At each point of the grid Zn is attached the same finite automaton, and such
a decorated point is called a cell. Each cell locally communicates with a finite
number of neighbors. The neighborhood is fixed and geometrically uniform. In this
paper we use the 8-neighborhood in dimension two and the 26-neighborhood in
dimension three. The neighbors of a point x = (x1, . . . , xn) ∈ Zn are the points
Int.
J. P
att.
Rec
ogn.
Art
if. I
ntel
l. 20
01.1
5:10
07-1
021.
Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
159
.84.
103.
173
on 0
3/14
/13.
For
per
sona
l use
onl
y.
October 25, 2001 9:23 WSPC/115-IJPRAI 00130
Generating Isotropic Discrete Waves on Cellular Automata 1009
where B = {(x1,1, . . . , xn,1), (x1,2, . . . , xn,2), . . . , (x1,|B|, . . . , xn,|B|)}.The local communications, which are deterministic and uniform, take place
synchronously according to discrete times.
Definition 2 (Configuration). A configuration CA of the cellular automaton Ais an application from Zn to S.
Definition 3 (Evolution). For all time t, the configuration CtA evolves at time
t+ 1 into the configuration Ct+1A defined by: x = (x1, . . . , xn) ∈ Zn and Ct+1
The function which associates the configuration Ct+1A to the configuration CtA
is called the global function of A.
A state q such that δ(q, . . . , q) = q is called a quiescent state. In the following,
we denote by I0 the initial configuration such that all the cells of the grid are in a
quiescent state except one cell, the center of the waves.
2.2. Signals
The natural way to describe the behavior of a cellular automaton is to exhibit all
the possible configurations of the cells during the evolution. When there are too
many possible configurations, it is better to conceive the succession of configurations
of this automaton, introducing its dynamic nature. To do this, we introduce the
notion of a signal.
Definition 4. At a given time t, a signal St on an n-dimensional cellular automa-
ton is a connected subset of cells {(x1(t), . . . , xn(t))/t ∈ N, t ≥ t0} that are in a
given subset of states of S.
In fact, the notion we use to describe the behavior of the cellular automata is not
exactly the signal but the “trace” of the signal along time. This leads to consider
surfaces in the space.
2.3. Construction of figures
We now clarify what building figures means in this paper. A figure is a finite subset
of Zn, and a family of figures is an application from N into the set of finite subsets
of Zn, that we denote F = (Fi)i∈N.
Constructing a figure F by a cellular automaton A is to select a subset SF of
A-states such that the automaton starting from a convenient initial configuration
reaches a configuration where a cell (x1, . . . , xn) is in a state belonging to SF if and
only if (x1, . . . , xn) ∈ F .
Definition 5 (Figures construction). Let A be an n-CA. We say that A con-
structs the family of figures F = (Fi)i∈N of Zn, according to the times (ti)i∈N if
Int.
J. P
att.
Rec
ogn.
Art
if. I
ntel
l. 20
01.1
5:10
07-1
021.
Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
159
.84.
103.
173
on 0
3/14
/13.
For
per
sona
l use
onl
y.
October 25, 2001 9:23 WSPC/115-IJPRAI 00130
1010 F. Feschet & L. Tougne
and only if there exists a subset SF and a sequence (ti)i∈N of times such that the
automaton starting from an initial configuration C0 at time t = 0 enters, at time
t = ti, a configuration where all the cells belonging to Fi are in a state of SF and
are the only ones to be in such states.
We are interested in constructing families of figures “as soon as possible”, that
means in real time.
Definition 6 (Real-time construction). Let A be an n-CA. We say that Aconstructs the family F in real time if and only if A constructs F according to the
times (i+ c)i∈N with c an arbitrary natural constant.
3. Strategy of Construction
3.1. From isotropy to anisotropy
The evolution of a cellular automaton is intrinsically local such that the construc-
tion of a discrete isotropic figure must be founded on local characterizations. The
fundamental idea is to proceed by duality, replacing the construction of a family of
isotropic figures by the construction of a family of anisotropic figures. For instance,
if we consider in dimension two the intersection between all the real concentric
circles and the two-dimensional grid, it is possible to prove that there are only
eleven different kinds of intersections (see Fig. 1).
In this figure, we see that starting from a cell of coordinates (x, 0) and continuing
on increasing y, the intersection remains constant and we reach something that
seems to be a “parabola” and, then it is always the same until we reach another
Fig. 1. We associate a color to each kind of intersection between the family of concentric circlesand the 2D-grid.
Generating Isotropic Discrete Waves on Cellular Automata 1015
Fig. 6. Examples of figures Ck.
The three-dimensional paraboloids Ck are obtained by the rotation of the two
dimensional ones (Hk) about the y-axis and Pk by the rotation about the z-axis
(see Fig. 6).
4.1. 2D construction
In this part, we summarized the work already done in Refs. 4 and 7. Following
Fig. 5, we can remark that the bundles of parabolas Hk and Vk separates the plane
into three parts as shown in Fig. 7.
The construction is done by mainly three automata. The first step consists in
generating Hk from one of its parts.
Property 1. There exists a CA which builds Hk from Hk ∩ (DV0 ∪K0).
The principle of this automaton is to use a diagonal signal which propagates
the diagonal movement state on a parabola (see Fig. 8).
As a corollary, we can construct all Hk from the intersection of the bundles Hand DV0 ∪ K0. The question to construct this intersection remains. Let us define
K0 = K01 ∪K02 by splitting K0 using the first diagonal and setting K01 the lower
part. Then,
DV0
DH0
K 0
K 02
K 01
Fig. 7. Decomposition of the plane: inDH0 the parabolas are horizontal, inDV0 they are vertical(symmetry of DH0) and in K0 this is the heart of the construction separated by symmetry intoK01 and K02.
Generating Isotropic Discrete Waves on Cellular Automata 1019
of such a property is the fact that if the radius of the circle grows on Ck, then it
grows on Pk′ , and if it is constant on Ck, it is constant on Pk′ .
Thus, it is possible to construct Pk′ from the family of Ck and by symmetry Ckfrom the family of Pk′ . We can remark that when k equals k′, the parabolas Hk
and Vk′ meet on the first diagonal and thus the radius of the circle which is the
intersection of P ′k and Ck is the same as the one of the circle at the intersection of
C′k and Pk.
To summarize, all the properties proven in dimension two are preserved in
dimension three and thus also is the existence of the cellular automaton used for
the construction.
5. Isotropic Waves
5.1. Construction of spheres from figures Ck and Pk
Let us suppose that we have already constructed the sphere of radius R and the
families Ck and Pk. The question that remains is to locally decide which cells belong
to the sphere of radius (R+ 1).
Let (x, y, z) be a cell such that there exists at least one cell in its 26-neighborhood
that belongs to the sphere of radius R. Then there are two cases that depend on
the position of the cell compared to the figures Ck:
• if the cell is on a figure Ck, then it belongs to the sphere of radius (R+1), because
the intersection between the sphere and a plane y = constant is the intersection
between a figure Ck and the same plane;
• if the cell is “inside” a figure Ck, then it belongs to the sphere of radius (R+ 1)
if and only if it has a 6-neighbor that belongs to the sphere of radius R.
Such rules are available for x ≥ y ≥ z ≥ 0, and symmetrical ones allow us to
obtain the spheres in all space. Figure 14 shows what we obtain for radii equal to
3, 5, 10 and 20.
Fig. 14. Examples of spheres (R = 3, 5, 10, 20).
5.2. What about time?
As the 2D-cellular automaton used to construct the circles generates the point of
coordinates (n, b√
2kn+ k2c) for every integer n and k at time (n+ k), we obtain
the intersection between the sphere and the plane y = n at time t = n+k: this is the
circle of center (0, n, 0) and of radius b√
2kn+ k2c. Consequently, if we consider the
sphere of radius R, its intersection by the plane y = R−k is obtained at time t = R.
So, the spheres are constructed in real time.
6. Conclusion and Future Works
In this article, we have presented a procedure to construct a family of concentric
spheres in real time on cellular automata. This extends a previous work.4 To do so,
we have defined a discrete sphere by its intersections with planes having one axis
x, y or z as normal vector. From this point of view, the work done for constructing
discrete circles can be used as for the generation of the discrete spheres. We have
presented a definition of the discrete sphere for the floor discretization, but any
well known discretization schemes in the literature can be used. For instance, it has
been proven that arithmetical circles can be constructed in real time, leading to a
new arithmetical sphere as compared to the one defined in Ref. 1. The presented
constructions remain essentially the same and we hope this can serve as a basis for
the generation of isotropic computations with cellular automata by using the duality
principle we propose between circles and parabolas, and spheres and paraboloids.
In future works, we plan to use this algorithm to the study of the intersections
between spheres in order to simulate wave interferences. This has applications in
physics and volume rendering. We also plan to study the consequences of a non-
integer center and/or noninteger radius for the presented work. The extension to
higher dimension is also of interest even if the principle of imposing the intersection
between an hypersphere of dimension n + 1 with an hyperplane to be an hyper-
sphere of dimension n will probably lead to a natural and recursive extension of
the proposed method.
References
1. E. Andres, Cercles Discrets et Rotations Discretes, Ph.D. Thesis, Universite LouisPasteur, Strasbourg, 1994.
2. E. Andres, “Discrete circles, rings and spheres,” Comput. Graph. 18, 5 (1994) 695–706.3. E. Andres and M.-A. Jacob, “The discrete analytical hyperspheres,” Trans. Visual.
Comput. Graph. 3, 1 (1997) 75–86.4. M. Delorme, J. Mazoyer and L. Tougne, “Discrete parabolas and circles on 2D cellular
automata,” Theoret. Comput. Sci. 218 (1999) 347–417.5. A. Kauffman, “Efficient algorithms for 3D scan conversion of parametric curves,
surfaces, volumes,” Comput. Graph. 4 (1987) 171–179.6. A. Kauffman and E. Shimony, “3D scan-conversion algorithms for voxel-based
graphics,” Proc. 1986 Workshop on Interactive 3D Graphics, eds. F. Crow andS. M. Pizer, October 1986, pp. 45–75.
7. L. Tougne, “Circle digitization and cellular automata,” Discrete Geometry forComputer Imagery (DGCI’96), Lecture Notes in Computer Science 1176, SpringerVerlag, pp. 283–294.
Int.
J. P
att.
Rec
ogn.
Art
if. I
ntel
l. 20
01.1
5:10
07-1
021.
Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
159
.84.
103.
173
on 0
3/14
/13.
For
per
sona
l use
onl
y.
October 25, 2001 9:23 WSPC/115-IJPRAI 00130
Generating Isotropic Discrete Waves on Cellular Automata 1021
Fabien Feschet re-ceived his Ph.D. in com-puter science from theInstitut National desSciences Appliquees inLyon in 1999. His re-search has been in medi-cal image analysis, med-ical image registrationprocedures for patient
positioning in cancer therapy and paral-lel processing. He is currently an AssistantProfessor in computer science in the LASSLaboratory of the University Lyon 1.
His main scientific interests lie in dis-crete geometry, general topology, operationresearch and image processing.
Laure Tougne receivedher Ph.D. degree incomputer science fromthe Ecole Normale Sup-erieure in 1997. Her re-search has been in thealgorithmic on 2D cel-lular automata. She is
currently an AssistantProfessor in computer
science in the ERIC Laboratory of the Uni-versity Lyon 2.
Her main scientific interests lie in discretegeometry.