Conference PaperPDF Available

New Gosper Space Filling Curves

  • January 2001
  • Conference: the International Conference on Computer Graphics and Imaging (CGIM2001),
Authors:
Hiroshi Fukuda at Kitasato University
Hiroshi Fukuda
Michio Shimizu
Michio Shimizu
G. Nakamura
G. Nakamura
G. Nakamura
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Download full-text PDF
Read full-text
Download citation
Content uploaded by Hiroshi Fukuda
Author content
All content in this area was uploaded by Hiroshi Fukuda
Content may be subject to copyright.
NEW GOSPER SPACE FILLING CURVES
HIROSHI FUKUDA MICHIO SHIMIZU GISAKU NAKAMURA
School of Administration and Informatics, Department of Liberal Art, Research Institute of Education,
University of Shizuoka, Nagano Prefecture College, Tokai University,
52-1 Yada, Shizuoka-shi, Shizuoka, 8-49-7 Miwa, Nagano-shi, Nagano, 2-28-4 Tomigaya, Shibuya-ku, Tokyo,
422-8526 JAPAN 380-8525 JAPAN 151-8677 JAPAN
ABSTRACT
We investigate several properties of a beautiful space
filling curve known as the Gosper curve, and try to
construct curves having the same properties. We found
17 new Gosper curves through computer searches, and an
infinite series of curves that are natural extensions of the
original Gosper curve.
KEY WORDS: Gosper curve, Space filling curves,
Fractal
1. INTRODUCTION
The Gosper curve is a space filling curve discovered by
William Gosper, an American computer scientist, in 1973,
and was introduced by Martin Gardner in 1976 [1,2]. The
Gosper curve is said to be a very beautiful and complex
monster curve because it has several special properties
that other space filling curves such as the Peano curve and
the dragon curve do not have.
In this paper, we point out several properties of the
Gosper curve and try to construct curves having the same
properties. Then, we show 17 new Gosper curves that we
found through computer searches. We also show an
infinite series of curves that are natural extensions of the
original Gosper curve.
In section 2, we describe the Gosper curve. In section 3,
we propose a construction method for new Gosper curves.
In section 4, results are shown and discussed.
2. GOSPER SPACE FILLING CURVES
The Gosper curve is a recursive curve constructed by
recursively replacing a dotted arrow, called the initiator,
by seven arrows, called generator, as shown in Fig. 1(a).
Fig. 1(b) and Fig. 1(c) illustrate the curves obtained by
replacing the initiator by generator once and twice
respectively.
(a) (b) (c)
Fig. 1 Gosper curve.
The arrows of generator in the curve obtained after
replacing the initiator to the generator any times are
located on the edge of a regular triangular lattice. The
degree of the generator arrow at the lattice point is two in
the interior of the Gosper curve and is one at the both
ends of initiator, i.e., the Gosper curve is the path from
the root to the tip of the initiator without any branches
visiting all interior lattice points. To our knowledge, the
Gosper curve is the only space filling curve to have these
properties. Therefore, in this paper we define the Gosper
curve in terms of these properties. We then try to find new
Gosper curves which satisfy our definition other than the
Gosper curve found by William Gosper. We call the
number of arrows included in the generator the size of
generator or Gosper curve itself.
3. COMPUTER SEARCH FOR NEW
GOSPER CURVES
We attach an equilateral triangle to the initiator arrow
and to the generator arrow so that the triangle has the
arrow as its edge and so that the direction of the arrow is
counterclockwise in the triangle as shown in Fig. 2 (a).
We call the equilateral triangle attached to each arrow a
“flag”.
(a) (b)
Fig. 2 Flags.
In the recursive procedure constructing Gosper curve,
the flag for the initiator is replaced by the seven flags for
the generator. The area of the flags is conserved during
the replacement. Further, the flags fill the triangular
lattice alternatively as shown in Fig. 2(b). From these
simple flag properties, the following procedure to
construct the generator of the Gosper curve can be
deduced.
(P1) Chose the size of the Gosper curve N.
(P2) Prepare an initiator arrow of length
L
and attach the
initial flag to it.
(P3) Prepare N flags with side length NL/ to be used
in the generator. The N flags should satisfy
conditions (P4) to (P7). Bellow, we call the set of N
flags “generator flags”.
(P4) The generator flags should be connected at their
vertices.
(P5) Two vertices of generator flags should coincide with
the ends of the initiator. From this condition the size
of Gosper curve is restricted to
xyyxN ++= 22 , ,...2,1,0, =yx (1)
Therefore the possible sizes of Gosper curves are
,...37,36,31,28,27,25,21,19,16,13,12,9,7,4,3=N(2)
(P6) The shape of the generator flags is symmetric under
a 3/2
π
rotation around its geometrical center.
(P7) If two coincident generator flags exist and one is
translated by a distance L in a direction parallel to
one of the three edges of a triangle in the initiator
flag, then the two generator flags should not overlap.
(P8) By taking only one edge from each flag and by using
the N edges of the generator flags, construct a path
without branches from the root to the tip of the
initiator.
(P9) Convert the edges in the path to the arrow so that the
direction of the arrow is counterclockwise in the flag.
We then obtain the generator arrow.
Using this procedure, we obtain the seven generator
flags shown in Fig.3 for 7≤N. The generator flags
which corresponds to the William Gosper’s curve is the
only generator flags satisfying conditions (P8) and (P9).
3=N 4=N 7=N
Fig. 3 Generator flags for 7≤N.
4. RESULTS AND DISCUSSIONS
As we have already mentioned, the smallest Gosper
curve obtained by the procedure given in the previous
section is the curve with 7=N found by William Gosper.
For 197
≤
<
N, there are two curves for 13
=
N
and 19
=
N. We show these curves together with their
generators and generator flags in Fig. 4 and Fig. 5,
respectively.
Fig. 4 A new Gosper curve for
N
= 13
Fig. 5 A new Gosper curve for
N
= 19.
For 3719
≤
<
N, there are seven and eight curves for
31
=
N and 37, respectively. The seven curves for
31
=
N are constructed by the three generator flags
shown in Fig.6 (31a), (31b) and (31c). Gosper curves
which have the same size can be distinguished by using
the flag and a serial number, for example (31a-1). It is
notable that five new Gosper curves can be produced
from the generator flags (31b). The curves produced from
the same generator flags have similar shapes.
(31a)
(31a-1)
(31b)
(31b-1)
(31b-2)
(31b-3)
(31b-4)
(31b-5)
(31c)
(31c-1)
Fig. 6 New Gosper curves for 31=N.
Eight curves for 37=N are constructed from the four
generator flags shown in Fig.7 (37a), (37b), (37c) and
(37d). The generator flags (37a), (37b), (37c) and (37d)
construct 1, 2, 3 and 2 curves, respectively. Although we
did not search for curves with 37>N because of the
limitation of our computer power, we expect that there
exist many big and complex Gosper curves.
(37a)
(37a-1)
(37b)
(37b-1)
(37b-2)
(37c)
(37c-1)
(37c-2)
(37c-3)
(37d)
(37d-1)
(37d-2)
Fig. 7 New Gosper curves for 37=N.
Next, we compare the three curves with sizes 7
=
N,
19=N and (37a-1). We found that the generator of the
curve for 19=N contains the generator of the curve for
7=N, and that the generator for (37a-1) contains the
generator for 19=N. Similarly, the generator flags of the
curve for 19=N are obtained by adding a layer of flags
around the generator flags of the curve for 7=N, and the
generator flags for (37a-1) are obtained by adding a layer
around the generator flags for 19
=
N. Thus the curve for
19=N and the curve (37a-1) are natural extensions of
the Gosper curve for 7=N.
It is possible to make the natural extension of the
Gosper curve for 7=N by adding a layer several times.
We explain the extension method as an example of the
curve for 37=N. First, we add a layer of flags around
the generator flags for 37=N as shown in Fig. 8. The
resulting generator consists of 61 flags. Thus we will
obtain the Gosper curve of the size 61=N. Next, we
choose the ends of initiator to be the leftmost vertex on
the button edge and the rightmost vertex, then we draw a
dotted line from the ends of the initiator to the ends of the
curve already drawn for 37=N as shown in Fig. 8.
After converting each segment of the dotted line to the
arrow, we obtain the generator of Gosper curve for
61
=
N. It is clear that bigger curves can be obtained by
continuing this procedure.
Fig. 8 Natural extension of the Gosper curve
for
N
= 61
As mentioned above, by adding n layers to the
generator flags for 7
=
N, we obtain a generator with size
133 2++= nnN ,...)3,2,1( =n. (3)
Thus we can obtain an infinite series of Gosper curve with
the size given by equation (3). The shape of the curves
obtained looks like snail.
In this paper, we pointed out several properties of the
monster curves found by William Gosper, and constructed
new monster curves having the same properties. We
found that the smallest size of the new monster curves is
13 and obtained an infinite series of natural extensions of
the Gosper curve. These new recursive monster curves are
beautiful and will be useful for computer graphics.
We should note that Gosper curve for 137
<
<N may
exist since the procedure (P1)—(P9) is not a necessary
and sufficient condition for our definition of the Gosper
curve. Finding a procedure corresponding to the necessary
and sufficient conditions and applying the procedure to
the other regular lattices are topics of future research.
REFERENCES
[1] M. Gardner, In which “monster” curves force
redefinition of the word “curve”, Scientific American
235 (December issue), 1976, 124-133.
[2] B. B. Mandelbrot, The fractal geometry, (New York ,
W. H. Freeman and Company, 1977).
... Each number in this trace is between 0 and S 2 − 1 where S is the scale factor for the fractal coordinate system, and identifies a sub-tile in bottom-to-top, left-to-right order, so that given within=tile coordinates (t x , t y ), the trace number n t = St y +t x . As a concrete example, the tile 1/(−1, −1) in the fractal coordinate system with scale 4 has trace coordinates 2/ [5]: it can be found by going up to the origin tile 2/(0, 0) and then descending into the sub-tile at (t x , t y ) = (1, 1) within that tile. Similarly, the tile 0/(2, 3) in scale 4 would have trace coordinates 2/ [14,3], whereas the tile 0/(2, 2) has trace coordinates 1/ [15]. ...
... It also generates this path incrementally: you can ask for any part of the path and it will give it to you without generating the rest of the path (although technically, it does generate some necessary constraints on the rest of the path in the process). In fact, these paths also form a family of space-filling curves [5,18] which incorporate a natural level of visual variety. (1) Each 5 × 5 region is traversed by a path that enters at one edge and exits at another, randomly chosen from all possible 5 × 5 paths which enter and exit at those coordinates (taking advantage of symmetries, there are less than two thousand such paths, and we simply store all of them in a cache indexable by entry/exit points). ...
... There are also threads of related work in mathematics (e.g., [5,15]), and physics (e.g., [15]) although this work does not directly apply any recent ideas from those fields. In particular, the discrete fractal coordinates presented here are far less complex than the notion of fractal coordinates present in [15]. ...
View
... Alternative space-filling fractal curves that map various other shapes. The Gosper curve, for instance, is often used to map larger hexagonal surfaces and its pattern is seen in Fig 7.4 [47]. ...
... Gosper fractal (space-filling) curve pattern[47] -increasing resolution images from left to right. ...
Improving image quality in holographic projections through novel iterative algorithmic solutions
Thesis
Full-text available
  • Jun 2023
Holographic displays have always shown great promise, a legitimate solution to current two dimensional stereoscopic displays. Holography is capable of demonstrating fully three-dimensional images without a loss in resolution, however, its current drawbacks such as comparatively limited field of view, limitations in hardware, with spatial light modulators mostly capable of modulating either the amplitude or phase of light, or its subpar image quality, have delayed its introduction and success in the mass market. This work focuses on the latter problem, finding novel algorithmic mechanisms for improving the image quality in replay fields. Given the nature of this issue within this field, there are numerous approaches for resolving this issue, more recently through machine learning, adopting supervised training models. This work follows a different approach. It investigates other industries, such as microscopy, astronomy, and computed tomography to draw parallels with the advances made there and adapted these implementations to computer generated holography. As a result, the Kaczmarz and later the Cimmino iterative algorithms were adapted to this problem and analysed extensively against Gerchberg-Saxton, a common algorithmic benchmark used in holographic projections. The results, in terms of image quality, are positive since in all cases these algorithms converge to a suitably low minima and display the intended replay field, whilst simultaneously outperforming Gerchberg-Saxton in several instances. Much like any other research proposal, there are always drawbacks, and trade-offs. Whilst these algorithms were shown to outperform the reference Gerchberg-Saxton in many scenarios, they are also comparatively computationally intensive and run significantly slower than the benchmark. Much of the work that followed under Cimmino was finding ways of combining these two attributes, i.e., improve the image quality as well as its speed. The Cimmino FFT algorithm was derived and equates to the more formally known Fienup algorithm. This work found a link between the Cimmino algorithm developed in the 1930s solving linear systems, to the Fienup algorithm developed in the 1970s, a well-known phase retrieval algorithm. Nevertheless, it was apparent when analysing the computational intensity of these algorithms the iterative use of discrete Fourier transforms (DFTs) and fast Fourier transforms (FFTs) requires significant processing power and is the bulk of the overall algorithmic run time. The final analysis of this thesis considers reducing this run time further by interpolating the replay field using a Hilbert curve. There is, however, another trade-off between the final image quality and the use of more DFT and FFTs against the use of interpolation, as expected. In conclusion, it can be verified that the larger the resolution of the image, the more benefit will be seen in terms of computational processing and algorithm run time, when adopting Hilbert interpolation.
View
... Since the early classics introduced by Peano, Hilbert, Polya, and Sierpinski, newer plane-filling curves have been discovered by several others, including Gosper [16], Mandelbrot [9], McKenna [10], Dekking [5], Fukuda [6], Ventrella [14], Arndt [1], Bandt [3], and others. As new curves are added to the repertoire, there is opportunity to make comparisons and identify common themes. ...
View
... The filling density of the covered area can be easily controlled by the order of the curve that is generated recursively like a fractal. A subset of SFCs, such as Hilbert, Gosper and Fukuda-Gosper [10] also possesses a self-avoidance capability. They forms a single closed path that will be interrupted in one or more parts in case a crack appears on its surface. ...
Stretchable Wireless Sensor Skin for the Surface Monitoring of Soft Objects
Conference Paper
  • Jul 2020
Like rigid objects, also soft and elastic manufactured materials for industrial and biomedical applications are subjected to fatigue stress that might speed up the aging process and even cause premature failures. The occurrence of early signs of damaging, like the arising of surface cracks, could avoid more severe critical events, especially when biomedical soft prosthesis are involved (such as artificial breast, stomach, bladder). A thin-film stretchable wireless sensor for surface monitoring is here proposed. The device is based on a densely distributed electrode exploiting, at the macro-scale, a Space-Filling Curve pattern, and a meandered profile in the micro-scale. Intercon-nection with a wrapped Radiofrequency Identification antenna permits to transmit the status of the electrode to remote, with no battery onboard. The device was manufactured by means of electron beam deposition over a thin elastomer. Surface defects of size larger than 0.9mm to 9mm can be detected with probability of 60% to 90%, respectively. Thanks to its double-scale meanderings, the sensor is highly tolerant to stretch keeping its shape nearly unchanged up to a 35% strain.
View
... Thus, for a given curve a resolution or space sensitivity can be defined. We consider hereafter the Gosper (N = 2) space filling curve [15] (Fig. 3). Accordingly, the space sensitivity is given by: To comply with the metal matter of the hip, generally made of titanium alloy, the RFID transponder to be connected with the crack electrode is a cavity-backed square loop as in [17]. ...
RF Detection of Micro-cracks in Orthopedic Implants by Conformal Space Filling Curves
Conference Paper
  • Sep 2019
Implanted prosthesis could be subjected to fractures due to defects and to aging. Conventional diagnostic tools involves X-Rays or, more commonly, the onset of the patient’s pain due to an irreversible failure. A non-invasive wireless monitoring system is here presented for the early detection of micro-cracks over metallic orthopedic implants. The proposed architecture involves a distributed electrode made of Space Filling Curves connected to the anti-tamper port of a UHF RFID transponder. The occurrence of an even small surface crack is detected in a binary form and transmitted remotely, outside the body following a standard RFID interrogation. The feasibility of the idea is supported by numerical analysis and experimental outcomes with a 3D printed and metallized hip prosthesis mockup. Preliminary results demonstrate a detection distance up to 0.7m, fully compliant with fast and non-collaborative diagnosis in the emerging Personalized Healthcare.
View
... A. Topological properties of SFC A SFC is univocally described in terms of initiator (d I ) and generator [32] (Fig. 13.a). The initiator is the starting path that coincides with the distance between the start and the end point of the curve and accordingly defines the size of the cell. ...
Space-Filling Electromagnetic Skins for the Wireless Monitoring of Surface Defects
Article
  • Aug 2019
Polymer-based objects (cable harness, gaskets, tires) are exposed, during their lifetime, to mechanical and chemical stress that often generates surface defects like crack and scratches. Early detection of signs of aging may enable a Predictive Maintenance to extend the life of the object and avoid severe failures. For this purpose, Space Filling Curves (SFC) are here proposed as an artificial electric skin, suitable to envelope a surface to wireless detect the presence of small aging signs by resorting to an electromagnetic backscattering platform. Size and resolution of the skin can be controlled by just two parameters and multiple skin cells can be arranged together to tessellate a large surface in order to even identify the position of the defect. By following a theoretical analysis of the sensor-oriented properties of SFCs, and in particular of the Gosper-Fukuda family, the feasibility of the idea is demonstrated by the way of preliminary experiments with a Radio Frequency Identification (RFID) IC, providing a 1-bit anti-tamper port.
View
Extraction of local structure information of point clouds through space-filling curve for semantic segmentation
Article
  • Nov 2022
The point cloud semantic segmentation network based on point-wise multi-layer perceptron (MLP) has gained extensive applications because of its end-to-end advantages. However, there exist two major limitations for this type of network: (1) In a neighborhood ball, the relative features between the central point and its neighboring point are not adequately exploited. (2) Only the relative features in the neighborhood ball are extracted, but the overall morphological information of the neighborhood ball is ignored. To overcome the limitations, this paper proposes two-fold improvements on local structure information extraction: (1) The relative angular feature is added and combined with other initial features as the input of the model to exploit the relative geometric features between points to the greatest extent. (2) The unordered points in the neighborhood ball are reordered based on the space-filling curve (SFC) and then fed into the MLP to extract the overall structure information of the neighborhood ball. And a semantic segmentation model is developed based on the two proposed feature extraction modules and U-Net, which is evaluated on two public datasets. The experimental result has demonstrated that the two proposed feature extraction modules can effectively extract geometric information in the point cloud and the proposed semantic segmentation model has strong semantic recognition capability for objects with complex morphologies. The mean intersection over union (mIoU) of the proposed model reached 70.6% and 47.8% for the Semantic 3D and Semantic KITTI datasets, respectively. Besides, the proposed model achieves real-time segmentation with only four encoder layers.
View
A Fractal-RFID Based Sensing Tattoo for the Early Detection of Cracks in Implanted Metal Prostheses
Article
  • Sep 2021
Manufacturing faults and aging are the main causes for micro-crack generation in implanted prostheses. An early detection of surface defects by means of local sensors can prevent dangerous complications and prompt for a preventive replacement of the medical device. For this purpose, a tattoo-like sensing mechanism based on pre-fractal Space Fulling Curves is wrapped onto the medical device and coupled with a zero-power RFID transponder. The resulting smart prosthesis is capable to identify the early formation of cracks and to communicate with the exterior of the body by backscattering communication. The crack detection method exploits the anti-tamper port of common Radiofrequency Identification (RFID) ICs and a small antenna, acting as harvester, closely integrated with the metal prosthesis. Simulations and tests with a mockup of metallic hip prosthesis and a leg phantom demonstrate that the device can identify surface cracks as small as 0.6 mm and can be wireless interrogated outside the body from up to 70 cm distance. The required geometrical change to the prosthesis is modest and does not hinder its mechanical robustness. Experiments also confirmed that the health status of the prosthesis could be even monitored on-the-fly when the patient crosses a door equipped with a UHF reader. The sensorized prosthesis could hence become an enabler for the emerging Precision Medicine and for the Internet of Bodies paradigm.
View
The generalized Sierpi\'{n}ski Arrowhead Curve
Article
  • Oct 2017
We define special Hamiltonian-paths and special permutations of the up-facing dark tiles on a checked triangular grid related to the generalized Sierpi\'{n}ski Gasket. Our definitions and observations make possible the generalization of the Sierpi\'{n}ski Arrowhead Curve for all orders. We produce these symmetric recursive curves in many ways by two kinds of asymmetric paths which are in a bijective relation and unambiguously transformable into each other in any order. These node-rewriting and edge-rewriting recursive curves keep their self-avoiding and simple properties after the transformation and their cardinality specifies a new integer sequence. We show a transformation table to change the curves into each other and we give another table to change them into Lindenmayer-system strings both by the absolute direction codes of their edges.
View
Sierpinski Curves
Chapter
  • Aug 2013
All space-filling curves discussed so far were based on a recursive substructuring into squares. The Sierpinski curve, in contrast, may be geometrically constructed using a recursive substructuring based on triangles. The curve is named after Waclaw Sierpinski, who – in 1912 – presented the respective mapping as the solution of certain functional equations. Similar to the Hilbert and Peano curve, the Sierpinski curve maps the unit interval onto a square.
View
The Fractal Geometry of Nature
Chapter
  • Mar 1983
"...a blend of erudition (fascinating and sometimes obscure historical minutiae abound), popularization (mathematical rigor is relegated to appendices) and exposition (the reader need have little knowledge of the fields involved) ...and the illustrations include many superb examples of computer graphics that are works of art in their own right." Nature
View
In which “monster” curves force redefinition of the word “curve”
  • 124-133
  • M Gardner
M. Gardner, In which “monster” curves force redefinition of the word “curve”, Scientific American 235 (December issue), 1976, 124-133.
The fractal geometry
  • Jan 1977
  • B B Mandelbrot
B. B. Mandelbrot, The fractal geometry, (New York, W. H. Freeman and Company, 1977).