Professional Documents
Culture Documents
An approach and an algorithm are introduced for circumscribing a given n-sided convex
polygon P,, by an m-sided polygon P,,,, (m < n), so that the added area is minimal. This
algorithm constitutes one building block in an algorithm for efficient nesting of arbitrary
geometric shapes in a given rectangular board. Flame cutting of steel sheets and laser cutting of
textiles are two industrial situations where this problem is of great importance. The approach
follows a top-down stepwise refinement and reduction of the original problem into simpler
subproblems, the solutions of all of which permit the solution of the original problem. It is first
shown that the optimal circumscription of P, by P, may be obtained by (n - m) iterative
single side reductions. The solution of the single side reduction problem is then characterized,
and an algorithm which is based on the triangle rotating side problem is proposed. This last
problem is concerned with passing the third side of a triangle through a given point that lies
within the area bounded by the two other given sides so that the triangle area is minimized. On
the way to proving the optimahty of the algorithms for the original problem and its
subproblems, new concepts and theorems are introduced. The algorithm was tested on a very
large number of polygons with varying numbers of sides and shapes, which were circumscribed
by hexagons. The average efficiency-defined as the ratio of the area of P, to that of Pm--was
96%. As n increases, efficiency reduces and approaches asymptotically the maximum achievable
efficiency for circumscribing a circle (“infinite” sided polygon) by a regular hexagon: 90.69%.
With n = 50 the average efficiency was 91.8%.
1. INTRODUCTION
In its classical form, the “cutting stock problem” is concerned with cutting a
rectangular board or roll of material into smaller rectangular pieces of given
different sizes so that the waste is minimized [l-3]. The problem has been presented
with various versions and limitations, such as restricting the cut to be of a
“guillotine” type [4,8] or tilting the rectangles relative to the board from which they
are cut [5]. None of the models, however, deals with shapes which are more complex
than rectangles, thereby severely restricting applicability. Flame and plasma cutting
of steel sheets and laser cutting of textiles are two instances of industrial applications
where optimal cutting of irregular shapes from a given rectangular two-dimensional
board is of great importance. Figure 1 illustrates figures nested on a board for flame
cutting of steel sheets.
The overall problem of our research deals with minimum waste nesting of
arbitrary shapes on a rectangular two-dimensional board. Since every shape can be
approximated, to any desired degree of accuracy, by polygons with sufficiently large
numbers of sides [lo], we chose to start with optimal nesting of polygons.
Optimal nesting of polygons has been previously addressed in the context of
paving, that is, which kind of polygons can be fitted side by side so that a continuous
*Address correspondence to M. Ben-Bassat.
131
0734-189X/83 $3.00
Copyright Q 1983 by Academic Press. Inc.
All rights of reproduction in any form reserved.
132 DORI AND BEN-BASSAT
covering of the plane is achieved without any waste at all. Kershner [6] has shown
that no convex polygon with more than six sides can pave the plane. He has also
shown that there are exactly three types of hexagons and eight types of pentagons
which can pave the plane, and has characterized the features of each type. (The first
three types of pentagons are special cases of the first three types of paving hexagons,
respectively.) Thus, the approach we have adopted for the efficient nesting of convex
polygons is based on finding the paving hexagon which circumscribes the originaI
convex polygon with minimum area addition.
Since paving hexagons have special features, we first find the circumscribing
general hexagon whose area is minimal, and then convert it to a paving hexagon with
minimal area addition.
This paper introduces an approach and a O(n) algorithm for optimal circumscrip-
tion (minimal waste) of an n-sided convex polygon by an m-sided polygon, where
n > m 2 3. The algorithm constitutes a part of the solution for the efficient nesting
of congruent shapes in a given rectangle, which, in turn, is a key component in the
cutting stock problem for irregular shapes.
Related works’ on this subject include an O(n) algorithm by Klee [ll] for finding
the minimum area circumscribing (MAC) triangle (m = 3) of a convex polygon P,,.
Freeman and Shapira [12] propose an algorithm for optimal circumscription of an
n-sided convex polygon by a rectangle, i.e., m = 4. Their algorithm is 0(n2).
Toussaint [13] shows how to reduce this complexity to O(n).
Dobkin and Snyder [14] discuss problems concerned with efficient inscription and
circumscription of convex polygons. In their paper they propose an O(n) algorithm
for an inscription problem which is dual to our problem, i.e., given P,,, what is the
largest Pk, k < n, which may be embedded in P,? The correctness of this algorithm
is proved, however, only for k = 3. Dobkin and Snyder [14, p. 131 also suggest
modifications to the inscription algorithm to make it applicable to the problem of
circumscribing a given P, by a triangle of minimal area.
‘We are grateful to an monynous referee for drawing our attention to related works of which we were
unaware.
CIRCUMSCRIBING A CONVEX POLYGON 133
A side 3 for which the above expression holds will be referred to as a “basic side,”
since intuitively it may serve as a stable basis on which the polygon can “stand”
firm. Geometrically, if we extend the two sides adjacent to a basic side, they will
form, together with the basic side, a triangle, which fully contains the original
polygon (see Figs. 2(b) and (c)).
(a) @I (cl
FIG. 2. The three kinds of convex polygons with more than three sides, classified by their basic order:
(a) circular polygon; (b) basic polygon of basic order 1: dA, + dA, i 27r; and (c) basic polygon of basic
order 2: dA, + LA, -C 277 and LA, + dA, < 2~.
134 DORI AND BEN-BASSAT
The basic order of a convex polygon P,, will be denoted by BO( P,,); thus,
THEOREM 1. Any two basic sides of a convex polygon P,, are adjacent.
According to Theorem 2 there are exactly three kinds of convex polygons with
more than three sides; one is circular and the other two are basic polygons; one is of
basic order 1, and the other is of basic order 2. The three kinds are illustrated in
Fig. 2.
4. PROBLEM REDUCTION
(1) Sj _ 1 and Sj are replaced by another side, Sj’, which passes through 5.;
(2) Sj-2 and $+I are extended until they meet Sj’;
(3) the remaining sides of P,-l coincide with those of P,,.
For a given J$, finding the new side s/‘, which yields minimum area addition, is a
constrained single side reduction problem which is formulated as follows (see Fig. 4).
Problem 2: For a given q of P,, find a side ?$’ passing through 5 such that
Solving Problem 2 requires the introduction of the “ triangle rotating side problem,”
which is stated as follows (see Fig. 3).
Problem 3: Given two fixed sides of a triangle and a point lying within the area
bounded by them, pass the third side through the given point so that the triangle
area is minimized.
THEOREM 6. The area of a triangle of which two sides, AB and AC are given, and
whose third side BC must pass through a given point P that lies within the area bounded
by AB and AC, is given by
sinx
sin(s;;x)u2 + 2uv +
sin(a+x)v2 (1)
where a is the angle between AB and AC,
x is the angle between AC and BC,
v is the distance from P to AB, and
u is the distance from P to AC (see Fig. 3).
*To simplify the notation, S,,, and Sje2 are also used to denote their extensions in any direction.
136 DORI AND BEN-BASSAT
dR 1 u2 u2
dx=--- 2 sin2x sin2(a + LX) ’
(2)
Equating dR/dx to zero we get
or
usina
X& = arctan
u - ucosa’
Since dR/dx changes its value from negative to positive as it passes through the
zero point, x is the point where R(x) attains its minimum. This minimal value is
determined by substituting (4) in (1):
R,,=+ (5)
sina
Additionally, from Pq. (3) we derive
V u
sin( x + a) = -sin x (6)
z, = 2lr - z,,
Z, = Aj + A,+1 - ?r,
and x as the angle between Sj’ and the continuation of Sj+ I (see Fig. 4). Using (4) we
can write
usin z,
xmin = arctan u - ucosz,
i 1
as the value of the angle x which ninimim the area of the triangle discussed above.
In our case, however, there is an additional constraint on x: ?$’ must lie completely
outside the area of the original polygon P,,, because otherwise P,- 1 will not
circumscribe P,. The range of existence of the area function R(x) (1) is therefore
reduced to
Z, < x I Aj+l. (7)
If xti is within these limits, then the constraint imposed by (7) is not effective. If, on
the other hand, xmin < ZX or xti > A,+l, then the optimal solution is constrained
to be xOrt = Z, or xopt = Aj+i respectively, so that 3’ will coincide with either S,-l
or Sj rather than cross the area of P,,. The optimal solution must be on these
boundary points since R(x) is monotonically decreasing up to xmin and monotoni-
cally increasing beyond it. Sj- i and Sj will be termed the “limiting sides.” The three
possible cases are demonstrated in Fig. 6.
The following theorem states the solution for external side reduction.
THEOREM 8. If Z, < 2m, then the optimal solution for the angle x is
z3 if xmin < Z,
xmin if Z, x xmin < A,,,
xopt =
Aj+l if xmin 2 Aj+i.
CIRCUMSCRIBING A CONVEX POLYGON 139
1 I
I
I
0
z3 *min A'J+l 22
R
J
I
I
,
I
0 2 A. LE.
‘min 3 ]+I
(IV, ?I,, q;,), because the area of the original polygon P,, plus the area of the
nonconvex quadrilateral (W, q-i, I$ y.,,), which is adjacent to it, is constant.
Thus, any increase of the area of the triangle (W, 71 i, y;,) causes a parallel
decrease of the area of the resulting P,,_ i in which VJis eliminated.
Analyzing the area function, we find that the area of the triangle increases as the
side S,‘, which passes through I$, tends to be parallel to either S,-2 or Sj+ ,. In our
case, however, since Sj’ must not cross the area of the original polygon (otherwise we
will not have a circumscribing polygon), the range of existence of the area function is
restricted to Z, I x I Aj+t, as in Case 1. Since the area of the triangle increases as
x approaches the boundaries, the optimal solution under these constraints will be
either Z, or A,+i, as stated in the following theorem.
where Cl1 = C(S,, S,-,) and V,;, = C(S,_,, S,,,) (see Fig. 5).
If Xopt = Z, then S; z Sj- i, or equivalently, S; passes through I$ and I$ i . If
X DPt= Aj+l then S; z Sj, or equivalently, 5” passes through I$ and V,,,.
Using Theorems 8 and 9, we can now solve the constrained single-side reduction
problem (Problem 2) for any given 7 to be eliminated. The algorithm proceeds by
first checking whether the original polygon is contained in the quadrilateral (Step 2).
If the answer is positive, we apply external side reduction (Theorem 8 in Steps 3 and
4), else we apply internal side reduction (Theorem 9 in Steps 5 and 6). In both cases
we compute and store the area Rj which is added to P, to yield Pnwl (Step 9).
Algorithm 2: Constrained Single Side Reduction for Vertex V, (CSSlX(j) (see Figs.
4, 5))
Step 1. Compute Z, = A, + Ajdl + A,+l.
step 2. If Z, r 2?~ go to Step 5, else continue.
Step 3. Compute Z,= 2?r - Z,,
usinz,
xti = arctan , and
i u - ucos z, I
Z3=A, + Aj+1 -2a.
Step 4. If xmin I Zs then xopt = Z,.
If Zs < xmin 5 A,,, then x,,r, = xmin.
If xmin > A,,, then xOrt = Aiil.
Go to Step 7.
Step 5. Find &=C(%., S,-,) and
~~,=c(sj-13 sj+,>3
CIRCUMSCRIBING A CONVEX POLYGON 141
~‘l=C(s,,,+,), and
I$l=c(s;,~+J.
Step 9. Compute and store the added area I; associated with the elimination of
5 as follows:
r,=R(C;,~-~,~‘,)+R(V,,~+,,~;,).
BO(P,n)
< BO(f-‘n) (8)
or
The following two theorems imply that (8) is impossible and that (9) can occur if
and only if at least one of the single side reductions involved in the iterative
circumscription is external
~EORBM 11. Any basic side which exists in P,, is retained as part of a basic side of
PIn‘
If the basic order of Pm = CirEpm( P,) is higher than that of P,,, it means that Pm
has b (b 5 3), basic sides, which did not exist in P,. Each such basic side is,
according to Theorem 12, a result of an external side reduction. External side
reduction may involve either the elongation of a side of P,, or the introduction of a
new side Sj’ which did not originally exist in P,,. If this new side passes through a
vertex which did not exist originally in P,, but was rather produced during the
iterative circumscription, then the new side is not tangent to P,. The resulting Pm in
this case is not optimal, because its area may be decreased by passing another side
parallel to S’ which is tangent to P,, (see Fig. 7).
To avoid a suboptimal result for Cir ’ - “( P,,), w h enever an external side reduction
is encountered during the iterative circumscription, the resulting new basic side must
be checked as to whether or not it passes through an original vertex of P,,. If the
answer is positive, then no improvement can be achieved, and the iterative cir-
cumscription may go on, else the basic side causes the elimination of a vertex which
CIRCUMSCRIBING A CONVEX POLYGON 143
FIG. 7. Compression.
was formed during one of the previous side reductions. In this case, the new basic
side must be moved until it is tangent to P,, such that the addition of area to P,, is
minimized. This process, which is described in the following algorithm, is termed
“COMPRESSION” and is optimal by construction.
Algorithm 3: Compression (see Fig. 7)
Step 1. Let V, denote the vertex which is eliminated by the new basic side sj’ and
letsi and sj+p denote the sides of P,, which intersect at V,.
Step 2. Set i = 0.
Step 3. Set i = i + 1.
Step 4. Perform external side reduction in which
(a) q+i is the fixed point,
(b) s/-l and %+,+i are the fixed sides,
(c) Sj+i-l and Sj+i are the limiting sides.
Step 5. Compute and store the area Ei added to the polygon by the addition of
the new side.
Step 6. If i = p continue, else go to 3.
Step 7. Find the minimal Ei and pass the new side through the corresponding
l$ +i. This side replaces 5’.
Step 8. End.
Step 1. Set i = 0.
Step 2. Set i = i + 1.
Step 3. Find P,,_ i = Cir( P, _ i + i) by a single side reduction (Algorithm 1).
Step 4. If the single side reduction, performed in Step 3, caused an addition of a
new basic side, then continue, else go to Step 6.
Step 5. If the new basic side does not pass through any one of the original
vertices of P, then perform compression (Algorithm 3), else continue.
Step 6. If i = n - m end, else go to Step 2.
Step 7. End.
The optimahty of this fktl algorithm is based on Theorem 4, which states that
Cir”-“( P,,) is the optimal P,,, if no new basic side is introduced during the iterative
circumscription, and on the optimality of the compression algorithm. The compres-
sion is applied whenever it becomes necessary, which makes sure that at the end of
each one of the (n - m) steps of the process, the resulting circumscribing polygon
P,,_ I is optimal not only with respect to P, _ i + i, but also with respect to P,.
It should be noted that for m 2 4, compression wiII never be applied more than
(2 - BO(P,)) times during the process of iterative circumscription of P,,, because,
according to Theorems 2 and 3, any convex polygon with more than three sides
contains at most two basic sides, and because according to Theorem 11, no basic
side which existed in P,, can be eliminated. Moreover, if there is one basic side (either
original or not), we know by Theorem 1 that the second basic side must be adjacent
to it. These facts facilitate the computational effort.
I I P
m+l
= Cir(Pm+2)
. I
SSR(n-m-l)
r---l
ip n-1
r 143 most_ 3 _CSP’~- _-__
-----~-- - - I- IL 1
I
SSR(i) I
Step 1. Select n angles from a uniform distribution on the interval (0,2n] and
sort them in an ascending order.
Step 2. Select n radii from a uniform distribution on (0, 11.
Step 3. Define the n vertices using the n angles and n radii.
Step 4. Check each three consecutive vertices of the resulting polygon for
convexity, and correct the middle one, if necessary, by lengthening its
radius by a proper random amount.
Step 5. Repeat Step 4 until no corrections are made over a complete cycle.
Step 6. End.
The result of this algorithm is an n-sided convex polygon which is confined within
a unit circle. The polygon of Fig. 7 is an example for n = 13.
Using the above simulation procedure for generating convex polygons, we applied
Algorithm 0 with m = 6 and n varying from 7 to 50. To avoid misleading results we
also tested polygons which are confined within an ellipsoid. This was achieved by
multiplying the x coordinates of the vertices by a certain number, termed “ovality.”
This number expresses the ratio between the major and the minor axis of the
circumscribing ellipsoid. For each n, four groups with oval&y of 1, 3,9, and 27 were
run. Each such group included 40 different random convex polygons.
146 DORI AND BEN-BASSAT
The results are presented in Table 1, where two lines contain the results for each n.
The first line summarizes the data for the 40 polygons with ovality 1 (i.e., contained
within the unit circle), while the second line summarizes the data for 120 polygons
with ovalities 3, 9, and 27. For example, for n = 10 the average circumscription
efficiency E(10,6) for ovality 1 is 98.71%, with standard deviation of 1.81% and
coefficient of variation (SD/E) 0.018. The average area of the polygons in this
group is 1.55, with standard deviation of 0.63 and coefficient of variation of 0.40.
TABLE 1
EWciency of Circumscribing n-Sided Polygons by Paver Hexagons
Average Average
Sample efficiency rhwn SD of
n size Ovality E(n,6) (73 PD[E(n,6)1 SD/E area area SD/area
For the 120 ten-sided polygons with ovalities 3, 9, and 27, the corresponding
numbers are 98.79%, 1.73%, 0.017, 18.28, 8.51, and 0.45.
Figure 9 demonstrates the circumscription efficiency E(n, 6) as a function of n.
The following conclusions may be drawn from the results presented in Table 1:
(1) It appears that the algorithm is not sensitive to variations in ovality.
(2) For the cases with ovality 1, as n increases, the circumscription efficiency
decreases monotonically from 99.8% for II = 7, to 91.8% for n = 50. This is very
close to the maximum achievable efficiency of 90.69% for a very large n as explained
below. Similar results hold for the cases with ovality greater than 1.
As n tends to infinity, the random convex polygon generated by our method
(Algorithm 4) tends to assume the shape of the circumscribing circle. Thus
lim R(P,)
n--too
= &. (12)
0’
5
‘0
E
al
t
2 96 -
5
i
E
a 95-
.$
%
s 94-
93 -
92 -
91
7 10 20 30 40 50
number of sides = n
i Ai = (n - 2)a. 64.1)
i=l
Suppose that S and S, are two nonadjacent basic sides in P,,, such that k # j f 1.
Denote
Zj=Aj+Aj+l CT (A4
and
Z, = A, + Ak+l -c T. (A4
Let
From (A.2) and (A.3) we conclude that Zj + Z, < 2~. Using this result together
with (A.l), we conclude that Z > (n - 4)~. By its definition, Z is a sum of (n - 4)
angles in a convex polygon, and as such, it cannot exceed (n - 4)77. This contradicts
the assumption that Sj and S, are not adjacent.
The left-hand side of (B.4) is the sum of 2(n - k) angles of a convex polygon,
hence it cannot exceed 2(n - k)n. This implies
which yields k < 4. Since k is the number of basic sides and hence must be an
integer, we conclude that BO(P,) I: 3, which proves the theorem.
APPENDIX C: PROOF OF THEOREM 3
We first note that a parallelogram has no basic side, a trapezoid has exactly one
basic side, and any other quadrilateral has exactly two basic sides, and never three.
Next, we assume that some convex polygon P,, with n 2 5 has three basic sides. By
Theorem 1, all three basic sides must be adjacent. Thus, they may be denoted by St,
S,, and S, without lack of generality. Using the definition of basic side we get
Al + A2 <?l
A, +A, < ?7
A, +A, < 77.
i A, = (n - 2)~. (C.2)
i==l
(C.4)
CIRCUMSCRIBING A CONVEX POLYGON 151
The left-hand side of (C.4) is the sum of n - 4 angles of a convex polygon, none
of which may exceed s, so we get a contradiction which proves our theorem.
and therefore
BC = CP + PB = -!-
sin x
+ ’
sin(x + u) . (E-1)
According to the sines theorem,
-=BC AC
sin a sin(n - x - u) .
AC=si4x+4 u + 0
sina i sin x sin(x + a) 1 *
h,=u+v. smx
sm(x+a).
Substituting (E-2) and (E.3) in the formula for triangle area R = AC - h,/2 we
obtain (l), which proves the theorem.
‘The proof of Theorem 12 is not based on any of the theorems presented in this paper.
152 DORI AND BEN-BASSAT
(a) By continuing S-i, T+i may be decomposed to q.1, and q.y i (see Fig.
F.l). T+i is a potential additional area for eliminating either I$ or I$+ i, but since T,
was selected before T; i, we conclude that lJ < T,> i, and therefore q < II+ i.
(b) Since T + I is a result of an external side reduction, it is composed of the two
triangles E,, i and E;+ i, the sum of which is q+ i (see Fig. F.2). By continuing S,,
E , + i is decomposed into E;;, i and Fi:i. As argued in (a), 4;,i 2 q, and since
1;;+1 -c Ei+l x Ti+l, we get T < 7;,,.
(c) By continuing Sj (see Fig. F.3), T+i is decomposed into Ei+l and E,I+,. As
argued in (a), E;,, 2 T, and since E,,, < q+i we get that q. < T+i.
The areas added to P4 in each one of the possibilities (1) (2) and (3) will be
denoted by R(l), R(2), and R(3), respectively. We will show that R(2) is smaller
than both R(1) and R(3).
“1
y------ --- “4 ---_ ,I
\ ----_
/ “3
1 S4 5 /
3 /
\
/
\“I : /
“3,
/’ \
/: K3
, ‘G, 5 tic,
/
/ \
, S? :
... \
/ 3
/ \
/ .... \
I,-‘_ ____________ .‘c’ ‘< .L----..- ------ Iv;
“I
“2
FIG. G.l. Replacing two nonbasic sides vs replacing two basic sides.
CIRCUMSCRIBING A CONVEX POLYGON 155
FIG. G.2. Replacing two nonbasic sides vs replacing one nonbasic and one basic side.
R(V;,h’,F/,)‘$W’;,&)
and
W) ’ ql).
156 DORI AND BEN-BASSAT
W) ’ $m
Passing the dotted lines (V,l, V;‘) and (Vi, V;‘) parallel to S, and S, respectively, we
get
and
and by (G.7)
Considering the fact that I( V,, Vr’) = I( V,, Vi) and that the corresponding angles
of the two quadrilaterals (V,, Vt’, V;l, V,) and (Vi, V,, V,, Vq’) are equal, then by
lengthening S, till G,, where its length equalizes the length of S,, and passing the
line (G,, G4) parallel to S,, we get a quadrilateral (Vi, V,, G,, G4) which is congruent
to (V,, V[, VF, V2). Since
R(3) ’ R(2),
FIG. G.3. Eliminating a basic side vs eliminating a nonbasic side in the case n L 5.
(b) n 2 5
Figure G.3 illustrates a situation where the following conditions are met:
(1) n = 5.
(2) St is a basic side, but S, and S, are nearly parallel.
(3) S, is relatively very short.
(4) The distance h of V, from S, is the same as its distance. from S,.
This is the worst case situation, because any possible change in one or more of the
above conditions causes an increase of the area R, added. to P,, when S, is eliminated
and/or a decrease of the area R, added to P, when S, is eliminated. We will show
that even in this case R, < R, which prevents the elimination of St.
(G-9)
and denote
and
I(&;, K) = c.
which implies
Since
(b:,6) <b+2c
we get
and
R, =R(I/,,V;,V,)+R(F:,V,,V~)
= $[2hb + 2h(b + c)] = h(2b + c).
and Ajil are external angles of the triangles (5, 5 ~, , v/I 1) and
(V”;lJ, J+,, y;,),
respectively, and as such, Ajpl = A;-.l + B and Aj+l = AJ+1 -i- B’.
By adding the two equations, we get
REFERENCES
1. P. C. Gilmore and R. E. Gomory, A linear programming approach to the cutting stock problem,
Oper. Res. 9, 1961, 849-859.
2. P. C. Gilmore and R. E. Gomory, Multistage cutting problems of two and more dimensions, Oper.
Res. 13, 1965, 94-120.
3. P. C. Gilmore and R. E. Gomory, The theory and computation of knapsack functions, Oper. Res. 15,
1967, 1045-1075.
4. N. Christofides and C. Whitlock, An algorithm for two dimensional cutting problems, Oper. Res. 25,
1977, 30-44.
5. P. De Cani, A note on the two dimensional rectangular cutting stock problem, J. Oper. Res. Sot.
Amer. 29 (7), 1978, 703-706.
6. R. B. Kershner, On paving the plane, Amer. Mu#b. Monthly 75, 1968, 839-844.
7. M. R. Spiegel, Mathematical Handbook of Formulas and Tables (Schaum’s Outline Series),
McGraw-Hill, New York, 1968.
8. A. Albano and R. Orsini, A heuristic solution of the rectangular cutting stock problem, Comput. J. 23,
1979, 338-343.
9. D. Dori and M. Ben-Bassat, Efficient nesting of congruent convex figures, Comm. Assoc. Comput.
Much ., in press.
10. J. Sklansky and V. Gonzalez, Fast Polygonal Approximation of Digitized Curves, Pattern Recognition,
August 1979.
11. V. Klee, “A linear-time algorithm that finds all local minima among triangles containing a given
convex polygon,” Proceedings, V. Symposium on Operations Research, Kijln, August 1980.
12. H. Freeman and R. Shapira, Determining the minimum area encasing rectangle for an arbitrary
closed curve, Comm. Assoc. Comput. Mach. 18, 1975,409-413.
13. G. T. Toussaint, “Pattern recognition and geometrical complexity,” 5th International Conference on
Pattern Recognition, pp. 1324-1347, Miami Beach, Fla., December 1979.
14. D. P. Dobkin and L. Snyder, “On a general method of m axinking and minimizing among certain
geometric problems,” Proceedings, 20th Symposium on Foundations of Computer Science, Puerto
Rico, October 1979.
15. L. Devroye, On the computer generation of random convex hulls, Comput. Math. Appl. 8,1982,1-13.