A model for digital topology

Gilles Bertrand and Michel Couprie

Abstract. In the framework known as digital topology, two di n erent

adjacency relations are used for structuring the discrete space Z . In this paper, we propose a model for digital topology based on the notion of order and discrete topology. We \validate" our model by considering the two fundamental notions of surface and simple point. At last, we give the di erent possible con gurations that may appear in 2- and 3dimensional surfaces in Z 4 which correspond to our model.

Keywords: discrete topology, digital topology, surface, simple point

Perhaps the simplest way for structuring the discrete space Z n is to consider two graphs, these graphs correspond to two adjacency relations between the elements of Z n: the direct and the indirect adjacency (for example, the 4- and the 8-adjacency in Z 2 ). This approach, known as digital topology 22, 12], is widely used in Z 2 for the applications to image analysis. Nevertheless, severe limitations appear in the discrete space Z 3 . For example, in the pioneering work of Morgenthaler and Rosenfeld 19], a de nition of surfaces in Z 3 is proposed, using such adjacency relations. This de nition is not based on structural properties which are fundamental for surfaces (e.g. the neighborhood of each point should constitute a \cycle" 8]). Hence, it is di cult to appreciate the relevancy of this de nition. Furthermore, this de nition seems to be almost impossible to generalize to higher dimensions. In this paper, we consider an approach based on the notion of order and discrete topology 1, 10, 14, 16]. We propose a model for digital topology by associating, to each subset of Z n , two orders: the hit (resp. miss) order corresponds to the indirect (resp. direct) adjacency. We \validate" our model by considering the two fundamental notions of surface and simple point. We show that the notion of surface in these two orders corresponds exactly to the notion of surface introduced by Morgenthaler and Rosenfeld. We also verify that the notion of simple point introduced in these two orders agrees with the corresponding notion in digital topology. These results are remarkable, since the de nitions of surface and simple point in the digital topology framework are fundamentaly di erent from ours. These

di erences explain why we used a computer in order to prove these results by an exhaustive checking. A hand-made proof would be of low interest. At last, we give the di erent possible con gurations that may appear in 2and 3-dimensional surfaces in Z 4 which correspond to our model.
2 Basic notions

In this section, we introduce some basic notions relative to orders (see also 6]). If X is a set, P (X ) denotes the set composed of all subsets of X , if S is a subset of X , S denotes the complement of S in X . If S is a subset of T , we write S T , the notation S T means S T and S 6= T . If is a map from P (X ) to P (X ), the dual of is the map from P (X ) to P (X ) such that, for each S X , (S ) = (S ). Let be a binary relation on X , i.e., a subset of X X . We also denote the map from X to P (X ) such that, for each x of X , (x) = fy 2 X; (x; y) 2 g. We de ne 2 as the binary relation 2 = n f(x; x); x 2 X g. An order is a pair jX j = (X; ) where X is a set and is a re exive, antisymmetric, and transitive binary relation on X . An element of X is also called a point. The set (x) is called the -adherence of x, if y 2 (x) we say that y is -adherent to x. Let (X; ) be an order. We denote the map from P (X ) to P (X ) such that, for each subset S of X , (S ) = f (x); x 2 S g, (S ) is called the -closure of S , (S ) is called the -interior of S . A subset S of X is -closed if S = (S ), S is -open if S = (S ). Let (X; ) be an order. We denote the relation = f(x; y); (y; x) 2 g, is the inverse of the relation . We denote = . The dual of the order (X; ) is the order (X; ). Note that (S ) = fx 2 S ; (x) S g, and (S ) = fx 2 S ; (x) S g. The set O composed of all -open subsets of X satis es the conditions for the family of open subsets of a topology, the same result holds for the set O composed of all -open subsets of X ; we denote respectively T = (X; O ) and T = (X; O ) these two topologies. These topologies are Alexandro topologies, i.e., topologies such that every intersection of open sets is open 1]. An order (X; ) is countable if X is countable, it is locally nite if, for each x 2 X , (x) is a nite set. A CF-order is a countable locally nite order. Let (X; ) be a CF-order. Let x0 and xk be two points of X . A path from x0 to xk is a sequence x0 ; x1 ; :::; xk of elements of X such that xi 2 (xi 1 ), with i = 1; :::; k. It may be seen that a CF-order (X; ) is connected for T (or for T ) if and only if it is path-connected, i.e, if for all x, y in X , there is a path from x to y. When (X; ) and (X 0 ; 0 ) are orders, a map f from X to X 0 is order preserving, or isotone, if y 2 (x) implies f (y) 2 0 f (x)]. We say that f is an isomorphism from (X; ) to (X 0 ; 0 ), if f is a bijection so that y 2 (y) if and only if f (y) 2 0 f (x)], for all x, y in X .

If (X; ) is an order and S is a subset of X , the order relative to S is the order jS j = (S; \ (S S )). Let (X; ) be an order. An element x such that 2 (x) = ; is said to be -terminal (for X ). A point y is an -terminal of x if y is an -terminal and y 2 (x). If (X; ) is an order, we de ne as the relation on X such that y 2 (x) if and only if y 2 2 (x) and 2 (x) \ 2 (y) = ;. The set (x) is called the -closeness of x, if y 2 (x) we say that y is -close to x. Let (X; ) be a CF-order and let x be a point of X . We say that x is unipolar if (x) consists in exactly one point. We say that x is -free if there is a sequence x0 ; :::; xk with xk = x, such that x0 is -unipolar and xi is -unipolar for the order jX nfx0 ; :::; xi 1 gj, i = 1; :::; k. A point which is not -free is called an -link. The -kernel of X is the subset of X composed of all -links of X . Two orders (X; ) and (X; 0 ) are said to be -equivalent if the orders induced by their -kernel are isomorphic. We use a general de nition for n-dimensional surfaces which has been proposed by Evako, Kopperman and Mukhin 7]. This notion is close to the notion of manifold used by Kovalevsky 17]; nevertheless it does not involve the necessity to attach a notion of dimension to each element of X , which allows to have a simpler de nition (in particular, no use of isomorphism is made). De nition 1: Let jX j = (X; ) be a non-empty CF-order. - The order jX j is a 0-surface if X is composed exactly of two points x and y such that y 62 (x) and x 62 (y). - The order jX j is an n-surface, n > 0, if jX j is connected and if, for each x in X , the order j 2 (x)j is an (n 1)-surface. - A (closed) curve is a 1-surface, a (closed) surface is a 2-surface. We will use the following notion of simple point as a model for the classical corresponding notion in Z 2 and Z 3 : De nition 2: Let jX j = (X; ) be a non-empty CF-order. - The order jX j is 0-contractible if X is composed of a single point. - A point x is n -simple if j 2 (x)j is (n 1)-contractible, n > 0. We denote X n the set composed of all points of X which are not n -simple points for jX j. - The order jX j is n-contractible, n > 0, if there is a sequence X 0 ; : : : ; X k with X 0 = X and X k = fag; a 2 X , such that X i = X i n1 if i is odd, and X i = X i n1 if i is even, for i = 1; : : : ; k.
3 Orders associated to

Zn

We give now a presentation of some orders which may be associated to Z n . Let E be a set and let E n be the Cartesian product of n copies of E . An element a of E n may be seen as a map from f1; :::; ng to E , a(i) is the i-th coordinate of a, i = 1; :::; n. If S is a subset of E n , the i-th projection of S is the set S (i) = fa(i); a 2 S g, i = 1; :::; n.

1, H 1, Let Z be the set of relative integers. We consider the families of sets H0 1 1 1 1 1 1. H such that, H0 = ffag; a 2 Zg, H1 = ffa; a + 1g; a 2 Zg, H = H0 H1 1 A subset S of Z n which is the Cartesian product of exactly m elements of H1 1 n n and (n m) elements of H0 is called a m-cube of Z . We denote H the set composed of all m-cubes of Z n , m = 0; :::; n. An m-cube of Z n is called a singleton if m = 0, a unit interval if m = 1, a unit square if m = 2, a unit cube if m = 3. In this paper, the basic order associated to Z n is the order (H n ; ), where = , thus y 2 (x) if x y. In Fig. 1, an example of a subset S of H 2 is given. The object S is made of two connected components S1 and S2 . It may be seen that S1 contains one -terminal and four -free elements (three -unipolar elements and one element which is an -unipolar element after the removal of these three elements). It follows that the -kernel of S1 is composed of a single element: jS1 j is -equivalent to an isolated point. In a similar way jS2 j is equivalent to an open curve made of three elements. 1

' $ r r r r r S ' $ ' $ r rg r r rg ' $ & % r r r r r & % & % S r r r r r & %

r r r r r

r r ' rg r rg r & r r r r

r r r $ ' $ r r r r rg r % & % r g r r r r r
(b)
H2

(a)

Fig. 1. A subset S of H 2 (a), and its dual in

(b).

In the framework of 2D-image analysis, an element of an object is called a pixel. In a square grid, a pixel is sometimes seen as an element of Z 2 and sometimes as an elementary square of R2 . In fact, there is an equivalence between the two approaches. This comes from the fact that there is a \geometrical" duality in a square grid: if we consider the four elementary squares which contain an element of Z 2 , these four squares may be considered in turn as \constituting a unit square". Note that we do not have such a duality in an hexagonal grid. In order to make more explicit this kind of duality, we propose the following construction. We de ne the dual of Z as the set Z = fp + 1=2; p 2 Zg. The dual of Z n is 1 , H 1 , H 1 such that, the set Z n = Z ]n . We consider the families of sets H0 1 1 1 1 1 1. H0 = ffag; a 2 Zg, H1 = ffa; a + 1g; a 2 Zg, H = H0 H1 n 1 A subset S of Z which is the Cartesian product of exactly m elements of H1

1 is called a m-cube of Z n , we denote H n the set and (n m) elements of H0 composed of all m-cubes of Z n, m = 0; :::; n. We denote f , the map from H n to H n , such that, for each x of H n , if x(i) = fp; p + 1g, then f (x)](i) = fp + 1=2g, and if x(i) = fpg, then f (x)](i) = fp 1=2; p + 1=2g, i = 1; :::; n. The dual of a m-cube of H n is the image under f of this m-cube, it may be seen that the image of a m-cube is a (n m)-cube. The dual of a subset S of H n is the subset of H n composed of the duals of all m-cubes in S . It may be seen that the map f is an isomorphism between the two orders (H n ; ) and ( H n ; ). In Fig. 1, the dual of a subset S of H 2 is represented. To conclude this section, we introduce some adjacency relations among elements of Z n which may be naturally derived from the order H n . Two elements p and q of Z n are said to be m-adjacent if there exists a m-cube of H n which contains both x and y. Two elements of Z n are strictly 0-adjacent if they are 0-adjacent (i.e. if they are equal), they are strictly m-adjacent if they are m-adjacent and not (m 1)-adjacent, with m = 1; :::; n.

The

i=d adjacency approach

In this section we give formal notions which correspond to the graph-theoretic approach introduced by Rosenfeld, this approach is based on two adjacencies relations, the i- and d-adjacencies. Two elements p and q of Z n are said to be indirectly adjacent or i-adjacent (resp. directly adjacent or d-adjacent) if they are n-adjacent (resp. 1-adjacent). It is well known that, in order to avoid connectivity paradoxes, we never use the same adjacency on both S and S . If we use the i-adjacency (resp. d-adjacency) for S , we must use the d-adjacency (resp. i-adjacency) for S . In the sequel the a-adjacency indicates the i- or the d-adjacency, i.e., we have a = i or a = d. Let p be an element of Z n . We denote a (p) the set composed of all elements of Z n which are a-adjacent to p. The a-neighborhood of p is the set a (p) = a (p) n fpg. An element q of a (p) is called a a-neighbor of p. For example, each p 2 Z 2 has 8 i-neighbors and 4 d-neighbors, each p 2 Z 3 has 26 i-neighbors and 6 d-neighbors. An a-path is a sequence p0 ; :::; pk , pi being a-adjacent to pi 1 , i = 1; :::; k. A subset S of Z n is a-connected if for all p, q of S there is an a-path from p to q which is composed solely of elements of S . A subset S of Z n is a simple closed a-curve if S is a-connected and if the aneighborhood of each element of S is made of exactly two elements. Let us consider the digital plane Z 2 . This set must be interpreted as a (nonbounded) surface. Thus it might be desirable that the neighborhood of each point of Z 2 could be interpreted by something like a simple closed curve. We see that the d-neighborhood of a point is made of four elements, and that any couple of these elements are not d-adjacent. In this sense the d-neighborhood of a point is \under-connected". In a similar way, we see that the i-neighborhood of a point is \over-connected". These peculiarities explain many of the di culties

encountered when using the i=d adjacency approach. Since the basic notion of a neighborhood is not appropriate, only ad-hoc de nitions are possible. For example, the notion of surface in Z 3 is far from being clear. Morgenthaler and Rosenfeld (MR) proposed the following de nition (see 19], 20], 21], 11]). Let (a; a) = (i; d) or (d; i). A MR a-surface is a nite connected subset S of Z 3 such that, for each p 2 S , we have: i) i (p) \ S ] has exactly one a-component a-adjacent to p; and ii) i (p) \ S ] has exactly two a-components which are a-adjacent to fpg, we denote C and D these components; and iii) 8q 2 a (p) \ S ], q is a-adjacent to both C and D. We conclude this section by giving a characterization of simple points in Z 3 ( 13, 5]). Let S Z 3 and p 2 Z 3 . Let (a; a) = (i; d) or (d; i). We denote 2 (p) = fq 2 Z 3 ; q is strictly 2-adjacent to pg. The geodesic a-neighborhood GS a (p) of p inside S is de ned by: GS (p) = i (p) \ S i GS d (p) = ( d (p) \ S ) fq 2 2 (p) \ S ; card( d (p) \ d (q ) \ S ) = 2g A point p 2 Z 3 is a-simple (for S ) if and only if the number of a-components in S GS a (p) and the number of a-components in Ga (p) both equal one.
5 The

h=m order approach

In order to build consistent topological notions for a subset S of Z n , we associate to S a subset (S ) of H n ; thus we recover the structure of a (discrete) topological space by considering the order (H n ; ). In this paper, the transformation is chosen in such a way that the induced topological notions may be seen as a \model" for the notions derived from the i=d adjacency framework. A natural idea for de ning is to consider \hit or miss" transformations 23]. Thus we consider the set S h composed of all elements of H n which have a non-empty intersection with S . In a dual way, we consider the set S m composed of all elements of H n which are included in S . Our model for the i=d adjacency approach consists in considering the -kernels of the two sets S h and S m as counterparts for the i- and d-adjacencies. De nition 3: Let S Z n. We denote S h = fx 2 H n; x \ S 6= ;g, and m S = fx 2 H n ; x S g. The h-set associated to S is the set h (S ) which is the -kernel of S h . The m-set associated to S is the set m (S ) which is the -kernel of S m . The h-order relative to S is the order ( h (S ); ), the m-order relative to S is the order ( m (S ); ), with = . We note that: - S m = S ]h ; - The set S h is -open, the set S m is -closed; - We have m (S ) = S m ; - The set h (S ) is, in general, neither -open nor -closed.

(a) (b): the h-order relative to S .

(b)

Fig. 2. (a): a subset S of Z 2 (the seven black dots) and the set Sh (black elements);
The proofs of the two following properties will be given in a forthcoming paper: Property 4: Let S Z n. The set h(S ) is composed of all elements fpg; p 2 S , and all m-cubes of Z n which contain two strictly m-adjacent points q and q0 of S , m = 1; :::; n. Property 5: Let S Z n. The dual of h(S ) is composed of the duals of all elements fpg; p 2 S , and all the m-cubes of Z n which are the intersection of the dual of fqg and the dual of fq0 g for some q; q0 2 S . In Fig. 2 (a), a set S h is represented. The original set S is represented by the seven black dots. We use the following conventions: a singleton is depicted by a circle ( ), a unit interval by a rectangle ( ), and a unit square by a square ( ). Two elements a, b are linked by a straight line if a 2 (b) or b 2 (a). In Fig. 2 (b), h (S ), the -kernel of the set S h , is given. It may be easily seen that: Property 6: Let S Z n. The set S is a closed curve for the h-order (resp. m-order) if and only if S is a closed i-curve (resp. closed d-curve). The following property has been proven by an exhaustive checking with the help of a computer: Property 7: Let S Z 3, let x be a point of S . The point x is i-simple if and only if fxg is 3 -simple for the h-order. The point x is d-simple if and only if fxg is 3 -simple for the m-order.
6 The

h-surfaces and m-surfaces in Z 3

Let S Z 3 , we say that S is an h-surface (resp. an m-surface) if S is a closed surface for the h-order (resp. m-order). In this section, we show that h-surfaces (resp. m-surfaces) are equivalent to the Morgenthaler and Rosenfeld (MR) i-surfaces (resp. d-surfaces) 19{21].

By construction, the MR's de nition cannot be extended in higher dimensional spaces. Furthermore, it may be asked whether it is possible to recover the fondamental structure of a combinatorial 2D-manifold ( 8]) where each element is \surrounded by a cycle". In fact an answer to this question has been given by Ayala and Al. (see 2], 3], 4]) in an original framework which provides a link between digital spaces and Euclidean spaces. A digital object is de ned as a digital manifold if its continuous analogue is a combinatorial manifold. In this framework, it has been shown that Morgenthaler's surfaces are combinatorial surfaces. In a previous work, Kong and Roscoe 11] had proposed a structural necessary and su cient condition based on speci c point con gurations, that characterizes the MR surfaces. Their approach is also based on continuous analogs. On the opposite, our approach is purely discrete. By drawing the inferences of the de nition of a 2-surface, we recover very simply the local point con gurations that may appear in h- or m-surfaces. We can even exhibit in the next section, the di erent con gurations that may appear in h-surfaces in Z 4 , as well as in 3-surfaces for h-orders in Z 4 . In the subsequent gures, we use the same conventions as for the twodimensional case, in addition, a unit cube is represented by a cube ( ). The following property is a direct consequence of Def. 1: Property 8: If S Z 3 is an m-surface, then each -terminal element x of m (S ) is a unit square, and 2 (x) is a closed curve composed of four singletons and four unit intervals (see b1 , Fig 3). Property 9: A subset S of Z 3 is a MR d-surface if and only if S is an m-surface.

b1

b2

b3

h-surface. The -terminal x is the front central unit square in b1 , the central unit cube in b2 and b3 .
In the sequel, we will use \utrs" as an abbreviation for \up to rotations and symmetries". Property 10: If S Z 3 is an h-surface, then i) for each -terminal element x of h (S ), 2 (x) corresponds (utrs) to one of the 3 con gurations bi (i = 1 : : : 3) depicted in Fig. 3, and

Fig. 3. The 3 possible con gurations for the set 2(x) of a -terminal x in an

ii) for each -terminal element x of h (S ), 2 (x) corresponds (utrs) to one of the 13 con gurations ai (i = 1 : : : 13) depicted in Fig. 4. The proof of this property is easily obtained by enumerating the di erent possible cases. It may be seen that each of the con gurations ai corresponds (utrs) to one of the con gurations that may appear in a MR i-surface (see for example 3], Fig. 3). We have indeed: Property 11: A subset S of Z 3 is a MR i-surface if and only if S is an h-surface.

a1

a2

a3

a4

a5

a6

a7

a8

a9

a10

a11

a12

a13

Fig. 4. The 13 possible con gurations for the set 2(x) of an -terminal x in an
h-surface. The -terminal x is the central singleton.

x y
y

(a)

(b)

Fig. 5. (a): a con guration in which y satis es the MR's de nition, but which may not appear in a MR i-surface; (b): the set 2 (y) for the h-order.

Remark: let us consider the con guration (a) in Fig. 5. We remark that y is an i-surface point, according to MR's de nition. But it is not possible that x satis es the conditions of the MR's de nition, whatever neighborhood it may have; in other words, the con guration (a) is not extensible (see 18] for a precise de nition of this notion). With Def. 1, the con guration (a) is directly rejected from the list of congurations that may appear in an h-surface: 2 (y) is not a closed curve (see Fig. 5 (b)).
7 The 2-surfaces and 3-surfaces for

h-orders in Z 4

In the Fig. 6, 7, the set (x), x being a 4-cube of Z 4 , is represented by three diagrams of 27 elements, s1 ; s2 ; s3 (from left to right). For each of these three diagrams, we use the same conventions as for the gures of the previous section, concerning the inclusion relation of the order (H 4 ; ). In addition, let us consider a in Fig. 6: each element x of s1 is included in x0 , the a! and ! the two vectors ! translated of x by a , and each element y of s3 is included in y0 , the translated of y ! by a . For example in Fig. 6 (c4 ), we have: x y; x z; x x0 ; z z 0 ; x0 z 0 . Property 12: If S Z 4 is an h-surface, then for each -terminal element x of h (S ), 2 (x) corresponds (utrs) to one of the 6 con gurations ci (i = 1 : : : 6): c1 ; c2 ; c3 correspond respectively to b1 ; b2; b3 depicted in Fig. 3, and c4 ; c5 ; c6 are depicted in Fig. 6. Property 13: If S Z 4 is a 3-surface for the h-order, then for each terminal element x of h (S ), 2 (x) corresponds (utrs) to one of the 5 con gurations di (i = 1 : : : 5) depicted in Fig. 7.
References

1. P. Alexandro , \Diskrete Raume", Mat. Sbornik, 2, pp. 501-518, 1937. 2. R. Ayala, E. Dom nguez, A.R. Frances, A. Quintero, J. Rubio, \On surfaces in digital topology", 5th Colloquium Discrete Geometry for Comp. Imagery, pp. 271276, 1995.

z
c4

z y a x y -a

c5

c6

Fig. 6. Three of the six possible con gurations for the set 2 (x) of a terminal x in 4
an h-surface in Z (the other three correspond to those in Fig. 3). 3. R. Ayala, E. Dom nguez, A.R. Frances, A. Quintero, \Determining the components of the complement of a digital (n 1)-manifold in Z n , Discrete Geometry for Computer Imagery, Vol. 1176, Lect. Notes in Comp. Science, Springer Verlag, pp. 163-176, 1996. 4. R. Ayala, E. Dom nguez, A.R. Frances, A. Quintero, \Digital lighting functions", Discrete Geometry for Computer Imagery, Vol. 1347, Lect. Notes in Comp. Science, Springer Verlag, pp. 139-150, 1997. 5. G. Bertrand, \Simple points, topological numbers and geodesic neighborhoods in cubic grids", Pattern Rec. Letters, Vol. 15, pp. 1003-1011, 1994. 6. G. Bertrand, \New notions for discrete topology", 8th Conf. on Discrete Geom. for Comp. Imag., Vol. 1568, Lect. Notes in Comp. Science, Springer Verlag, pp. 216-226, 1999. 7. A.V. Evako, R. Kopperman, Y.V. Mukhin, \Dimensional Properties of Graphs and Digital Spaces", Jour. of Math. Imaging and Vision, 6, pp. 109-119, 1996. 8. J. Francon, \Discrete combinatorial surfaces", CVGIP: Graphical Models and Image Processing, Vol. 57, pp. 20-26, 1995. 9. J. Francon, \On recent trends in discrete geometry in computer science", Conf. on Discrete Geometry for Comp. Imag., pp. 3-16, 1996.

10. E. Khalimsky, R. Kopperman, P. R. Meyer, \Computer Graphics and Connected Topologies on Finite Ordered Sets", Topology and its Applications, 36, pp. 1-17, 1990. 11. T. Y. Kong, A. W. Roscoe, \Continuous Analogs of Axiomatized Digital Surfaces", Computer Vision, Graphics, and Image Processing, 29, pp. 60-86, 1985. 12. T.Y. Kong and A. Rosenfeld, \Digital Topology: introduction and survey", Comp. Vision, Graphics and Image Proc., 48, pp. 357-393, 1989. 13. T.Y. Kong, \Topology-Preserving Deletion of 1's from 2-, 3- and 4-Dimensional Binary Images", Conf. on Discrete Geometry for Comp. Imag., Lect. Notes in Comp. Science, Vol. 1347, Springer Verlag, pp. 3-18, 1997. 14. T. Y. Kong, R. Kopperman, P. R. Meyer, \A Topological Approach to Digital Topology", American Mathematical Monthly, 38, pp. 901-917, 1991. 15. R. Kopperman, P.R. Meyer and R.G. Wilson, \A Jordan surface theorem for threedimensional digital spaces", Discrete Comput. Geom., 6, pp. 155-161, 1991. 16. V. A. Kovalevsky, \Finite Topology as Applied to Image Analysis", Computer Vision, Graphics, and Image Processing, 46, pp. 141-161, 1989. 17. V.A. Kovalevsky, \Topological foundations of shape analysis", in Shape in Pictures, NATO ASI Series, Series F, Vol. 126, pp. 21-36, 1994. 18. R. Malgouyres, \A de nition of surfaces of Z 3 ", Conf. on Discrete Geometry for Comp. Imag., pp. 23-34, 1993. See also Doctoral dissertation, Universite d'Auvergne, France, 1994. 19. D.G. Morgenthaler and A. Rosenfeld, \Surfaces in three-dimensional images", Information and Control, 51, 227-247, 1981. 20. G.M. Reed and A. Rosenfeld, \Recognition of surfaces in three-dimensional digital images", Information and Control, Vol. 53, pp. 108-120, 1982. 21. G.M. Reed, \On the characterization of simple closed surfaces in three-dimensional digital images", Computer Vision, Graphics and Image Proc., Vol. 25, pp. 226-235, 1984. 22. A. Rosenfeld, \Digital Topology", Amer. Math. Monthly, 86, pp. 621-630, 1979. 23. J. Serra, Image Analysis and Mathematical Morphology, Academic Press, 1982.

y
d1

d2

d3

d4

d5

for h-orders in Z . This set constitutes a 2-surface (a closed surface). For three elements p = y; z; t, the set 2 (p) is represented with bold lines. This set constitutes a 1-surface (a closed curve).

Fig. 7. The 5 possible con gurations for the set 2 (x) of a terminal x in a 3-surface 4