Ingeniería y Competitividad ISSN: 0123-3033 [email protected] Universidad del Valle Colombia Cardona, Luis F.; Múnera, Luis E. Self-Similarity of Space Filling Curves Ingeniería y Competitividad, vol. 18, núm. 2, 2016, pp. 113-124 Universidad del Valle Cali, Colombia Available in: http://www.redalyc.org/articulo.oa?id=291346311010 How to cite Complete issue Scientific Information System More information about this article Network of Scientific Journals from Latin America, the Caribbean, Spain and Portugal Journal's homepage in redalyc.org Non-profit academic project, developed under the open access initiative Ingeniería y Competitividad, Volumen 18, No. 2, p. 113 - 124 (2016) COMPUTATIONAL SCIENCE AND ENGINEERING Self-Similarity of Space Filling Curves INGENIERÍA DE SISTEMAS Y COMPUTACIÓN Auto-similaridad de las Space Filling Curves Luis F. Cardona*, Luis E. Múnera** *Industrial Engineering, University of Louisville. KY, USA. ** ICT Department, School of Engineering, Department of Information and Telecommunication Technologies, Faculty of Engineering, Universidad Icesi. Cali, Colombia. [email protected]*, [email protected]** (Recibido: Noviembre 04 de 2015 – Aceptado: Abril 05 de 2016) Abstract We define exact self-similarity of Space Filling Curves on the plane. For that purpose, we adapt the general definition of exact self-similarity on sets, a typical property of fractals, to the specific characteristics of discrete approximations of Space Filling Curves. We also develop an algorithm to test exact self- similarity of discrete approximations of Space Filling Curves on the plane. In addition, we use our algorithm to determine exact self-similarity of discrete approximations of four of the most representative Space Filling Curves. We found that SFCs like Moore’s based on recursive structure are actually not self- similar, highlighting the need to establish a formal definition of the concept for SFCs. Keywords: Fractals, self-similarity, Space-Filling Curves. Resumen El propósito de este artículo es desarrollar un test que permita determinar la auto-similaridad de una Space Filling Curve (SFC), estudiándolas desde el punto de vista de la teoría fractal y concentrándonos en la propiedad de auto-similaridad. El test consiste de dos fases, en la primera se identifica una partición especial de la curva denominada partición CM y luego se muestra que la curva es auto-similar si y sólo si es auto-similar bajo dicha partición. Adicionalmente, el test es aplicado a cuatro famosas SFC (Peano, Moore, Meander y Lebesgue) para determinar su auto-similaridad. Se encuentra que algunas SFC como la de Moore con estructura recursiva y aparente auto-similaridad son en realidad no auto-similares, resaltando la necesidad de formalizar el concepto. Palabras Clave: Auto-similaridad, fractales, Space Filling Curves. 113 Ingeniería y Competitividad, Volumen 18, No. 2, p. 113 - 124 (2016) 1. Introduction pattern recognition in several contexts (Rosenthal & Kouri, 1973; Lam & Liu, 1996). In 1878, Cantor established the equipotency between R and any n-dimensional euclidean In the literature of Space Filling Curves, space Rn (Sagan, 1994a). In the plane, it implies mathematicians have stated a close relation the existence of a bijective mapping from R to between SFCs and fractals. SFCs have patrons and R2 (an interval can be mapped into the plane). regularities that also are commonly found in fractals A year later, Netto proved that such mapping and that relation has been linked to the self-similarity cannot be continuous (Bader, 2013). A decade property of fractals (Plaza et al., 2005; Haverkort & later, Peano proved the existence of a continuous Walderveen, 2010; Skubalska-Rafajowicz, 2005). surjective mapping from R to R2 (Sagan, 1994c). Wunderlich (1954) linked fractals and SFCs by He proposed a continuous curve that goes through introducing Iterated Functions System (IFS) in every point of the plane (Figure 1), and those the generation of discrete approximation of Space curves were called Space Filling Curves (SFCs). Filling Curves. IFS is a method to generate discrete SFCs has fascinated mathematicians since then, approximation of fractals by means of their self- and brilliant mathematicians as Hilbert, Moore, similarity property and recursive algorithms. In Lebesgue, Meander, Poyla and Sierpinski has this way, a discrete approximation of a fractal of proposed their own SFCs exploring different nth order is composed of a finite number of copies properties and perspectives (Sagan, 1994a). of itself that can only differ from the original in size and orientation and are integrated under a regular patron consistent through iterations. McClure (2003) studied the self-similarity of the Hilbert curve, and lays out the advantages derived from this property by utilizing Iterated Functions System (IFS) as Wunderlich (Hutchinson, 1981). But, since not all SFCs are exact self-similar, in this paper, we propose a method to determine whether a discrete approximation of Space Filling Curve is exact self-similar. Figure 1. A discrete approximation of the Despite the extensive attention that SFCs have Peano Space Filling Curve. received in the literature and their evident link with fractals, there is no formal definition But mathematicians are not the only ones who for their self-similarity. This paper presents a have been fascinated with SFCs, they also formal conceptualization of the self-similarity are extensively used in engineering (Bader, adapted to the context SFCs and an efficient 2013). SFCs are used to solve combinatorial computational test to determine whether a optimization problems in industrial contexts specific SFC is self-similar. (Hu & Wang, 2004; Hua & Zhou, 2008). They also have been used as a scanning technique for The paper is organized as follows: Section 2 image processing and compression (Dafner et al., presents an introduction to the theory of self- 2000). Anotaipaiboom & Makhanov (2005) used similarity and to SFCs theory, also a definition of SFCs to plan movement paths of robotic arms exact self-similarity of discrete approximations with numeric control. Additionally, Valle & Ortiz of SFC. Section 3 presents the test that determine (2011), Bahi et al. (2008), and Chen & Chang whether a discrete approximation of a Space (2005) have explored applications of SFCs for Filling Curve on the plane is exact self-similar. In clustering. Furthermore, SFCs have been used for Section 4, we determine self-similarity of four 114 Ingeniería y Competitividad, Volumen 18, No. 2, p. 113 - 124 (2016) of the most representative SFCs using our test. Finally, as in geometry a triangle is represented by three Section 5 provides conclusions and future research. non-collinear points, we represent a SFC by means of an ordered finite list of points, SFC = 2. Context {p1,p2,……,pn}. Figure 3a shows the Hilbert curve and the circles indicate the set of points by By definition, every fractal has the property of self- which it is represented. In this case, the Hilbert similarity (Barnsley, 1993). A set is self-similar if curve is found in the 4x4 lattice, hence 16 points when we examine small portions of it, we only see are used to represent it. The order in which these reproductions of our initial set (Rubiano, 2002), as points are found in the list, is the same order in shown in Figure 2. which these points are found in the curve. Figures 3b, 3c, 3d and 3e show partitions of the curve of Figure 3a. Figure 3b is a partition with four equivalence classes, and each equivalence class has four points. Note that the union of the four equivalence classes equals the 16 points in Figure 3a and no point belongs to more than one equivalence class. Figure 2. Sierpinski triangle. Ferrari (2007) establishes a formal definition of self-similarity, as follows: Definition 1: 2 A compact set X ∩ R is self-similar if there is a finite Figure 3. Different cardinality partitions of Hilbert curve. set of contractive∩ similarities FN ={F1,F2,…Fn} such n that X= i=1 Fi (X) and the subsets of X of the form Fi (X) overlap only in the boundaries. 3. Self-Similarity Test Contractive similarities are affine transformations It is seen in Figure 2 that the largest triangle is of the form F = r D x+b , where r is a scalar such i i i i i made up of three smaller triangles which in turn that 0< r < 1, D is the matrix of rotations ∩ and b is i i i are made up of three smaller ones. To prove self- a vector in R2 (Barnsley 1993). Note that n F (X) i=1 i similarity of this set, it is sufficient to show itsIFS , is a cover of X. In addition, the intersection of which is easily found by inspection. However, it is subsets F (X) is only allowed on the borders, a i not always evident to prove self-similarity of a set. property which makes it a quasi-partition. In SFCs context, the latter restriction is even stronger, According to the definition of self-similarity, in since these curves go through every point ∩ of the order to prove that a set is not self-similar, it is plane exactly once. Thus, in SFCs context, n F (X) i=1 i necessary to show the non-existence of a partition is a partition of X and every Fi (X) is an such that all equivalence classes p of be equivalence class. i contractive similarities of X, that is, In this paper, we analyze SFCs from a combinatorial point of view, just as Dai & Su (2003) do. As well 115 Ingeniería y Competitividad, Volumen 18, No. 2, p. 113 - 124 (2016) Thus the formula may be written as, of partitions included in H until the only one left is CM. The basic foundations for the process are the following two propositions: This would imply that to prove the self-similarity Proposition 1: It is necessary that all equivalence of a compact set X, it is necessary to evaluate classes pi of be affine transformations of X, all partitions of X and find in everyone of them for X to be self-similar under .