Egyptian Informatics Journal (2012) 13, 147–154

Cairo University

Egyptian Informatics Journal

www.elsevier.com/locate/eij
www.sciencedirect.com

ORIGINAL ARTICLE

Shape preserving rational bi-cubic function

Malik Zawwar Hussain a, Maria Hussain b,*
, Madiha Amjad a

a
Department of Mathematics, University of the Punjab, Lahore, Pakistan
b
Lahore College for Women University, Lahore, Pakistan

Received 13 October 2011; revised 26 June 2012; accepted 28 June 2012

Available online 10 October 2012

KEYWORDS Abstract The study is dedicated to the development of shape preserving interpolation scheme for
Rational bi-cubic functions; monotone and convex data. A rational bi-cubic function with parameters is used for interpolation.
Parameters; To preserve the shape of monotone and convex data, the simple data dependent constraints are
Monotone surface; developed on these parameters in each rectangular patch. The developed scheme of this paper is
Convex surface confined, cheap to run and produce smooth surfaces.
 2012 Faculty of Computers and Information, Cairo University.
Production and hosting by Elsevier B.V. All rights reserved.

1. Introduction A fine quantity of work [1–9] has been published in previ-

ous years that emphasized on the shape preservation of curves
Monotonicity is a key tool for the specification of Digital to and surfaces. Beatson and Ziegler [2] developed a shape pre-
Analog Converters (DACs), Analog to Digital Converters serving interpolation scheme for regular monotone data. The
(ADCs) and sensors. These devices are enormously used in rectangular patches were joined by C1 quadratic spline. Suffi-
control system applications. Erythrocyte sedimentation rate cient conditions were derived on data values and derivatives
(ESR) in cancer patients, uric acid level in patients suffering arranged on the rectangular grid to ensure monotonicity. Carl-
from gout, approximation of couples and quasi couples in sta- son and Fritsch [3] developed a bi-cubic polynomial interpola-
tistics are a few other monotone quantities. Convexity arises in tion scheme for shape preservation of monotone data. The
the data generated in nonlinear programming, scientific appli- conditions were worked out on derivatives. These conditions
cations such as design, optimal control, parameter estimation were sufficient to establish monotone bi-cubic polynomial over
and approximations. all rectangular elements. Sarfraz, Butt and Hussain [8] con-
structed a rational interpolation scheme for monotone data.
The scheme was cubic in each co-ordinate parameter. The ra-
tional interpolant had one free parameter in each subinterval.
* Corresponding author. Tel.: +92 0331 4930071.
Monotonicity was assured by establishing automotive con-
E-mail addresses: [email protected] (M.Z. Hussain),
[email protected], [email protected] (M. Hussain).
straints on the free parameter.
Peer review under responsibility of Faculty of Computers and
Asaturyan [1] proposed a subdivision scheme for convex
Information, Cairo University. surface data interpolation. The rectangular patches in which
convexity was lost were identified and divided into nine sub-
rectangles. Convexity preserving interpolation scheme was
applied on these sub-rectangles. It was observed that change
Production and hosting by Elsevier
in a sub-rectangle affected the whole domain thus established
1110-8665  2012 Faculty of Computers and Information, Cairo University. Production and hosting by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.eij.2012.06.001
148 M.Z. Hussain et al.
 
it a global scheme. Costantini and Fontanella [4] developed an diþ1  di diþ1  di
vi ¼ wi ¼ qi þ Max ; ; qi ¼ 0:
interpolation scheme for regular convex data. The scheme was Di  di diþ1  Di
semi global. Floater [5] derived sufficient conditions on the
The proofs of Theorem 1 and Theorem 2 can found in [9].
control points of Bézier surfaces to raise convex surface from
convex data. Hussain, Sarfraz and Shaikh [7] developed a C2
rational scheme for shape preservation of positive and convex 3. Rational bi-cubic function
data.
The paper underlines the problem of monotone and convex
In this section, the extension of rational cubic function (1) to
data interpolation. The data values are interpolated by ra-
the rational bi-cubic function S(x, y) for the interpolation of
tional bi-cubic function. This rational bi-cubic function enjoys
regular data is described. Let the regular data be arranged over
eight parameters in each rectangular patch. The range of these
the rectangular domain D ¼ ½a; b  ½c; d. The partition of the
parameters is determined to ensure shape preservation of data.
intervals ½a; b and ½c; d are p : a ¼ x0 < x1 < x2 <   
The developed schemes of this paper deal positively to both
< xm ¼ b and p ~ : c ¼ y0 < y1 < y2 <    < yn ¼ d. The repre-
data and data with derivatives, assure local command on the
sentation of rational bi-cubic function over the rectangular
surface and promise C1 smoothness.
patch I ¼ ½xi ; xiþ1   ½yj ; yjþ1  is
The rest of the paper is organized as follows: The Section 2
reviews [9]. Section 3 details the rational bi-cubic function [6] Sðx; yÞ ¼ a0 ðaÞSðxi ; yÞ þ a1 ðaÞSðxiþ1 ; yÞ  Sðx; yj Þb0 ðbÞ
used for surface data interpolation. The shape preserving inter-  Sðx; yjþ1 Þb1 ðbÞ
polation schemes for monotone and convex data are developed  
 a0 ðaÞ b0 ðbÞSðxi ; yj Þ þ b1 ðbÞSðxi ; yjþ1 Þ
in Section 4. Section 5 concludes the paper.  
 a1 ðaÞ b0 ðbÞSðxiþ1 ; yj Þ þ b1 ðbÞSðxiþ1 ; yjþ1 Þ ; ð2Þ
2. Rational cubic function ai(a) and bi(b), i = 0, 1 are the cubic Hermite blending func-
tions defined as:
This section details the rational cubic function developed by a0 ðaÞ ¼ ð1  aÞ2 ð1 þ 2aÞ; a1 ðaÞ ¼ a2 ð3  2aÞ; b0 ðbÞ
Sarfraz and Hussain [9]. x  xi
Let the planar data under consideration be fðxi ; fi Þ : ¼ ð1  bÞ2 ð1 þ 2bÞ; b1 ðbÞ ¼ b2 ð3  2bÞ; a ¼ ;b
di
i ¼ 0; 1; 2; . . . ; ng and the partition of knot be xi < xiþ1 ; y  yj
i ¼ 0; 1; 2; . . . ; n  1. The function and derivative values ¼ ; i ¼ 0; 1; 2;    ; m  1; j ¼ 0; 1; 2;    ; n  1:
^
dj
are fi, di respectively. A piecewise C1 rational cubic function
is defined as: Sðx; yj Þ; Sðx; yjþ1 Þ; Sðxi ; yÞ and Sðxiþ1 ; yÞ are the rational cubic
p ðhÞ function (1) constituting the four boundaries of rectangular
SðxÞ  Si ðxÞ ¼ i ; ð1Þ patch I ¼ ½xi ; xiþ1   ½yj ; yjþ1  as:
qi ðhÞ
X 3
where ð1  aÞ3i ai s0;i
i¼0
pi ðhÞ ¼ Ui ð1  hÞ3 þ vi Vi hð1  hÞ2 þ wi Wi h2 ð1  hÞ þ Zi h3 ; Sðx; yj Þ ¼ ; ð3Þ
v1 ðaÞ
with
qi ðhÞ ¼ ð1  hÞ3 þ vi hð1  hÞ2 þ wi h2 ð1  hÞ þ h3 ;
s0;0 ¼ Fi;j ; s0;1 ¼ vi;j Fi;j þ di Fxi;j ; s0;2 ¼ wi;j Fiþ1;j  di Fxiþ1;j ; s0;3
hi di hi diþ1 ¼ Fiþ1;j ; v1 ðaÞ
Ui ¼ fi ; Zi ¼ fiþ1 ; Vi ¼ fi þ ; Wi ¼ fiþ1  ;
vi wi
¼ ð1  aÞ3 þ vi;j að1  aÞ2 þ wi;j a2 ð1  aÞ þ a3 ;
x  xi fiþ1  fi
h¼ ; hi ¼ xiþ1  xi ; Di ¼ : X
3
hi hi ð1  aÞ3i ai s1;i
vi and wi are the free parameters, variations to the values of vi’s i¼0
Sðx; yjþ1 Þ ¼ ; ð4Þ
and wi’s help the user to control (tighten or loosen) the shape v2 ðaÞ
of the curve in different pieces. One can note that when, with
vi = wi = 3, then the rational cubic function obviously be- s1;0 ¼ Fi;jþ1 ; s1;1 ¼ vi;jþ1 Fi;jþ1 þ di Fxi;jþ1 ; s1;2
come cubic Hermite interpolant.
¼ wi;jþ1 Fiþ1;jþ1  di Fxiþ1;jþ1 ; s1;3 ¼ Fiþ1;jþ1 ; v2 ðaÞ
Theorem 1 [9]. For the given conditions di P 0; i ¼ 0; 1; 2; ¼ ð1  aÞ3 þ vi;jþ1 að1  aÞ2 þ wi;jþ1 a2 ð1  aÞ þ a3 ;
. . . ; n on the derivative parameters, the sufficient conditions for
the interpolant (1) to be monotonically increasing are X
3
ð1  bÞ3i bi s2:i
i¼0
di þ diþ1 di þ diþ1 Sðxi ; yÞ ¼ ; ð5Þ
vi ¼ ; wi ¼ : v3 ðbÞ
Di Di
with
Theorem 2 [9]. For the given conditions d1 < D1 <    s2;0 ¼ Fi;j ; s2;1 ¼ v^i;j Fi;j þ ^ ^i;j Fi;jþ1  ^
dj Fyi;j ; s2;2 ¼ w dj Fyi;jþ1 ; s2;3
< Di1 < di < Di <    < Dn1 < dn on the derivative parame-
¼ Fi;jþ1 ; v3 ðbÞ
ters and data, the sufficient conditions for the interpolant (1)
to be convex are ¼ ð1  bÞ3 þ v^i;j bð1  bÞ2 þ w
^i;j b2 ð1  bÞ þ b3 ;
Shape preserving rational bi-cubic function 149
( )
X
3
Fyiþ1;j þ Fyiþ1;jþ1
ð1  bÞ3i bi s3;i v^iþ1;j ¼ w
^iþ1;j > Max 0; :
i¼0
b iþ1;j
D
Sðxiþ1 ; yÞ ¼ ; ð6Þ
v4 ðbÞ
Proof. Let fðxi ; yj ; Fi;j Þ : i ¼ 0; 1; 2; . . . ; m; j ¼ 0; 1; 2; . . . ; ng be
with
the monotone regular data. The data is arranged over the rect-
s3;0 ¼ Fiþ1;j ; s3;1 ¼ v^iþ1;j Fiþ1;j þ ^dj Fy ; s3;2
iþ1;j angular region I ¼ ½xi ; xiþ1   ½yj ; yjþ1 ; i ¼ 0; 1; 2; . . . ; m  1;
^iþ1;j Fiþ1;jþ1  ^dj Fyiþ1;jþ1 ; s3;3 ¼ Fiþ1;jþ1 ; v4 ðbÞ
¼w j ¼ 0; 1; 2; . . . ; n  1. Since the data is monotone so
Fxi;j > 0; Fyi;j > 0; Di;j > 0 and D b i;j > 0, (the necessary condi-
¼ ð1  bÞ3 þ v^iþ1;j bð1  bÞ2 þ w
^iþ1;j b2 ð1  bÞ þ b3 : tions for monotonicity of data).
The monotonicity of rational function (2) is dependent on
3.1. Arithmetic mean method of derivative approximation [6] the monotonicity of boundary curves ((3)-(6)). The boundary
curves are monotone if Sð1Þ ðxl ; yÞ > 0 and Sð1Þ ðx; yk Þ > 0;
In this paper the partial derivatives Fxi;j and Fyi;j are approxi- l ¼ i; i þ 1 and k ¼ j; j þ 1. The complete expressions of
mated by ‘arithmetic mean method.’ Arithmetic mean method Sð1Þ ðxl ; yÞ > 0 and Sð1Þ ðx; yk Þ > 0 are as follows:
of derivative approximation is a difference scheme which in-
volves three neighbouring points for the computation of deriv- X
5

atives at each ijth corner of rectangular patch. The explicit ð1  aÞ5i ai s4;i
i¼0
formulae are as follows: Sð1Þ ðx; yj Þ ¼ ; ð7Þ
ðv1 ðaÞÞ2
d0
Fx0;j ¼ D0;j þ ðD0;j  D1;j Þ ;
d0 þ d1 with
s4;0 ¼ Fxi;j ; s4;1 ¼ 2ðwi;j Di;j  Fxiþ1;j Þ þ Fxi;j ;
dm1
Fxm;j ¼ Dm1;j þ ðDm1;j  Dm2;j Þ ;
dm1 þ dm2
s4;2 ¼ Di;j þ 2ðwi;j Di;j  Fxiþ1;j Þ þ ðvi;j wi;j Di;j  wi;j Fxi;j  vi;j Fxiþ1;j Þ;
Di;j þ Di1;j
Fxi;j ¼ ; i ¼ 1; 2; 3;    ; m  1; j ¼ 0; 1; 2;    ; n:
2 s4;3 ¼ 3Di;j þ 2ðwi;j Di;j  Fxi;j Þ þ ðvi;j wi;j Di;j  wi;j Fxi;j  vi;j Fxiþ1;j Þ;

^d0 s4;4 ¼ 2ðvi;j Di;j  Fxi;j Þ þ Fxiþ1;j ; s4;5 ¼ Fxiþ1;j :

b i;0 þ ð D
Fyi;0 ¼ D b i;0  D
b i;1 Þ ;
^d0 þ ^d1
The Sð1Þ ðx; yjþ1 Þ can be computed by replacing j P by j + 1 in (7).
^dn1 The boundary curve Sðx; yj Þ is monotone if 5i¼0 ð1  aÞ5i
b i;n1 þ ð D
Fyi;n ¼ D b i;n1  D
b i;n2 Þ ; ai s4;i > 0, which is true only if s4;i ’s are positive. s4,i’s, i = 0, 1,
^dn1 þ d^n2
2, 3, 4, 5 are positive if the parameters vi;j and wi;j observe the
following restrictions
b i;j þ D
D b i;j1
Fyi;j ¼ ; i ¼ 0; 1; 2;    ; m; j ¼ 1; 2; 3;    ; n  1;  
2 Fxi;j þ Fxiþ1;j
F vi;j ¼ wi;j > Max 0;
F
where Di;j ¼ iþ1;j i;j and D b i;j ¼ Fi;jþ1^Fi;j . Di;j
di dj

Similarly, the boundary curve Sðx; yjþ1 Þ is monotone if

4. The proposed algorithms  
Fxi;jþ1 þ Fxiþ1;jþ1
vi;jþ1 ¼ wi;jþ1 > Max 0; :
Di;jþ1
In this section the monotonicity and convexity preserving
schemes are developed for regular surface data using the ra- The computations for monotonicity of boundary curves
tional bi-cubic function (2). Sðxl ; yÞ; l ¼ i; i þ 1 are

4.1. Monotone rational bi-cubic function X

5
ð1  bÞ5i bi s5;i
i¼0
Sð1Þ ðxi ; yÞ ¼ ; ð8Þ
Theorem 3. The rational function defined in (2) preserves the ðv3 ðbÞÞ2
shape of monotone data if in every rectangular patch, the free with
parameters vi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; v^i;j ; w ^iþ1;j satisfy the
^i;j ; v^iþ1;j ; w
s5;0 ¼ Fyi;j ; s5;1 ¼ 2ðw b i;j  Fy Þ þ Fyi;j ;
^i;j D
following conditions i;jþ1

  b i;j þ 2ðw b i;j  Fy Þ þ ð^ b i;j  w

Fxi;j þ Fxiþ1;j s5;2 ¼ D ^i;j D i;jþ1 ^i;j D
vi;j w ^i;j Fyi;j  v^i;j Fyi;jþ1 Þ;
vi;j ¼ wi;j > Max 0; ;
Di;j
( ) b i;j þ 2ðw
s5;3 ¼ 3 D b i;j  Fyi;j Þ þ ð^
^i;j D b i;j  w
^i;j D
vi;j w ^i;j Fyi;j  v^i;j Fyi;jþ1 Þ;
Fyi;j þ Fyi;jþ1
v^i;j ¼ w
^i;j > Max 0; ;
b i;j
D
s5;4 ¼ 2ð^ b i;j  Fyi;j Þ þ Fy ; s5;5 ¼ Fy :;
vi;j D i;jþ1 i;jþ1
 
Fxi;jþ1 þ Fxiþ1;jþ1 Similarly, for the expression of Sð1Þ ðxiþ1 ; yÞ, replace i by i + 1
vi;jþ1 ¼ wi;jþ1 > Max 0; ;
Di;jþ1 in (8).
150 M.Z. Hussain et al.
P5 5i i
S(xi, y) is monotone if i¼0 ð1  bÞ b s5;i > 0, which is
true only if s5,i’s are positive. s5,i’s, i = 0, 1, 2, 3, 4, 5 are
positive if the parameters v^i;j and w ^i;j observe the following
restrictions
( )
Fyi;j þ Fyi;jþ1
v^i;j ¼ w
^i;j > Max 0; :
b i;j
D
Similarly, the boundary curve S(xi+1, y) is monotone if
( )
Fyiþ1;j þ Fyiþ1;jþ1
v^iþ1;j ¼ w
^iþ1;j > Max 0; :
b iþ1;j
D

Algorithm 1. It is the algorithm to apply Theorem 3.

Step 1: Compute ðm þ 1Þ  ðn þ 1Þ monotone data points

fðxi ; y j ; F i;j Þ : i ¼ 0; 1; 2; . . . ; m; j ¼ 0; 1; 2; . . . ; ng.
Step 2: Approximate the derivatives F xi;j and F yi;j at knots
using Section 3.1.
Step 3: Determine the values of free parameters
vi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; ^vi;j ; w
^ i;j ; ^viþ1;j ; w
^ iþ1;j by using
Theorem 3. Figure 2 xz-view of Fig. 1.
Step 4: Insert the values of F i;j ; F xi;j ; F yi;j ; i ¼ 0; 1; 2;
. . . ; m; j ¼ 0; 1; 2; . . . ; n and vi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; ^vi;j ;
w ^ iþ1;j ; i ¼ 0; 1; 2; . . . ; m  1; j ¼ 0; 1; 2; . . . ; n  1
^ i;j ; ^viþ1;j ; w
in rational bi-cubic function (2).

4.1.1. Demonstration
In this section two monotone data sets are considered. These
data sets are interpolated by bi-cubic Hermite (Figs. 1–3 and
7–9) and monotonicity preserving scheme (Figs. 4–6 and 10–
12) developed in Section 4.1. It is observed that bi-cubic Her-
mite although C1 is unable to preserve the monotone shape of
data, whereas, the monotonicity preserving scheme of Sec-
tion 4.1 preserves the shape of data.

Example 1. The surface in Fig. 1 represents the bi-cubic

Hermite interpolation of monotone data considered in Table 1.

Figure 3 yz-view of Fig. 1.

The xz-view and yz-view of this figure is provided in Figs. 2

and 3 respectively. It is clear from Fig. 3 that the bi-cubic
Hermite loses the monotone shape of data for y e [100, 200].
The same data set is interpolated in Fig. 4 by C1 monotone
rational bi-cubic function developed in Section 4.1. Figs. 4–6
establish that monotonicity preserving scheme (developed in
Section 4.1) works well.

Example 2. The Table 2 is of 3D data set, monotone over the

whole domain.
The monotone data set in Table 2 is interpolated by bi-cubic
Hermite in Fig. 7 which loses the monotone shape of data. It is
Figure 1 Bi-cubic Hermite interpolation. more clear in Figs. 8 and 9 which provide the xz and yz views of
Shape preserving rational bi-cubic function 151

Figure 4 C1 monotone rational bi-cubic function.

Figure 6 yz-view of Fig. 4.

Table 1 A 3D monotone data set.

y/x 1 100 200 300
1 0.1354 0.1744 0.3632 5.3814
100 13.5400 17.4370 36.3230 538.1400
200 27.0800 34.8730 72.6450 1076.3000
300 40.6200 52.3100 108.9700 1614.4000

Table 2 A 3D monotone data set.

y/x 0.1 0.3 0.7 1.3
0.1 0.0570 0.0572 0.0579 0.0593
0.3 0.0795 0.0801 0.0829 0.0869
0.7 0.6932 0.1451 0.1601 0.1715
1.3 0.1382 13.8160 15.8950 17.1111

( )
Figure 5 xz-view of Fig. 4. Fyi;jþ1  Fyi;j Fyi;jþ1  Fyi;j
v^i;j ¼ w
^i;j > Max 0; ; ;
b i;j  Fyi;j Fy  D
D b i;j
same figure. Fig. 10 is the interpolation of same data by C1 i;jþ1

monotone rational bi-cubic function. It is comprehensible from ( )

Fig. 10 that C1 monotone rational bi-cubic function developed Fxiþ1;jþ1  Fxi;jþ1 Fxiþ1;jþ1  Fxi;jþ1
vi;jþ1 ¼ wi;jþ1 > Max 0; ; ;
in Section 4.1 preserved the monotone shape of data. Di;jþ1  Fxi;jþ1 Fxiþ1;jþ1  Di;jþ1
( )
4.2. Convex rational bi-cubic function Fyiþ1;jþ1  Fyiþ1;j Fyiþ1;jþ1  Fyiþ1;j
v^iþ1;j ¼ w
^iþ1;j > Max 0; ; :
b iþ1;j  Fy
D Fy Db iþ1;j
iþ1;j iþ1;jþ1

Theorem 4. The bi-cubic function defined in (2) preserves the

Proof. Let {(xi, yj, Fi,j): i = 0, 1, 2, . . . , m; j = 0, 1, 2, . . . , n}
convexity if in each rectangular patch I = [xi, xi+1] · [yj, yj+1]
be the given data arranged over the rectangular grid and sat-
parameters vi,j, wi,j, vi,j+1, wi,j+1, v^i;j ; w ^iþ1;j satisfy
^i;j ; v^iþ1;j ; w
isfy the following necessary conditions of convexity
the following conditions
( ) b i;j 6 D
Di;j 6 Diþ1;j ; D b i;jþ1 ; Fx 6 Fx ; Fyi;j 6 Fy ; Fx
Fxiþ1;j  Fxi;j Fxiþ1;j  Fxi;j i;j iþ1;j i;jþ1 i;j
vi;j ¼ wi;j > Max 0; ; ; b i;j 6 Fy :
Di;j  Fxi;j Fxiþ1;j  Di;j 6 Di;j 6 Fxiþ1;j ; Fyi;j 6 D i;jþ1 ð9Þ
152 M.Z. Hussain et al.

Figure 7 Bi-cubic Hermite interpolation.

Figure 9 yz-view of Fig. 7.

Figure 8 xz-view of Fig. 7. Figure 10 C1 monotone rational bi-cubic function.

over each rectangular patch I = [xi, xi+1] · [yj, yj+1], i = 0, 1, with

2, . . . , m  1; j = 0, 1, 2, . . . , n  1. s6;0 ¼ s4;1  ð2vi;j  1Þs4;0 ;
The rational function (2) interpolates convex data as
convex surface for convex boundary curves ((3)–(6)). The s6;1 ¼ 2s4;2  ðvi;j  2Þs4;1  ðvi;j þ 4wi;j Þs4;0 ;
convex boundary curves are obtained by computing the
constraints on the parameters ðvi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; v^i;j ; s6;2 ¼ 3ðs4;3 þ s4;2 Þ  3wi;j s4;1  3ðwi;j þ 2Þs4;0 ;
^iþ1;j Þ in the following way.
^i;j ; v^iþ1;j ; w
w
s6;3 ¼ 4s4;4 þ 4ðvi;j þ 1Þs4;3 þ ðvi;j  2wi;j Þs4;2  ð2wi;j þ 5Þs4;1
The boundary curves S(xl, y) and S(x, yk), l = i, i + 1;
k = j, j + 1 are convex if S(2)(xl, y) > 0 and S(2)(x, yk) > 0,  5s4;0 ;
l = i, i + 1; k = j, j + 1.
s6;4 ¼ 5s4;5 þ ð2vi;j þ 5Þs4;4 þ ð2vi;j  wi;j Þs4;3  ðwi;j  4Þs4;2  4s4;1 ;
P7 7i i
i¼0 ð1  aÞ a s6;i
Sð2Þ ðx; yj Þ ¼ ; ð10Þ
di ðv1 ðaÞÞ3 s6;5 ¼ 3ðvi;j þ 2Þs4;5 þ 3vi;j s4;4  3ðs4;3 þ s4;2 Þ;
Shape preserving rational bi-cubic function 153

s6;6 ¼ ð4vi;j þ wi;j Þs4;5 þ ðwi;j  2Þs4;4  2s4;3 ;

s6;7 ¼ ð2wi;j  1Þs4;5  s4;4 :

The S(2)(x, yj+1) can be obtained by replacing j by j + 1 in
(10).
P7The boundary
7i i
curve
P7 S(x, yj)
7i i
is convex if
i¼0 ð1  hÞ h s6;i > 0. i¼0 ð1  hÞ h s6;i > 0 if s6,i > 0,
i = 0, 1, 2, . . . , 7.
s6,i > 0, i = 0, 1, 2, . . . , 7 if
( )
Fxiþ1;j  Fxi;j Fxiþ1;j  Fxi;j
vi;j ¼ wi;j > Max 0; ; :
Di;j  Fxi;j Fxiþ1;j  Di;j

Similarly, it can established that the boundary curve

S(x, yj+1) is convex if the parameters vi,j+1 and wi,j+1 satisfy
the following constraints
( )
Fxiþ1;jþ1  Fxi;jþ1 Fxiþ1;jþ1  Fxi;jþ1
vi;jþ1 ¼ wi;jþ1 > Max 0; ; :
Di;jþ1  Fxi;jþ1 Fxiþ1;jþ1  Di;jþ1

In the same way, for the convexity of other two vertical

Figure 12 xz-view of Figure 11.
boundary curves, we have
P7 with
7i i
i¼0 ð1  bÞ b s7;i
s7;0 ¼ s5;1  ð2^
vi;j  1Þs5;0 ;
Sð2Þ ðxi ; yÞ ¼ ; ð11Þ
^di ðv ðbÞÞ3
3
s7;1 ¼ 2s5;2  ð^
vi;j  2Þs5;1  ð^
vi;j þ 4w
^i;j Þs5;0 ;

s7;2 ¼ 3ðs5;3 þ s5;2 Þ  3w

^i;j s5;1  3ðw
^i;j þ 2Þs5;0 ;
Table 3 A 3D convex data set.
y/x 3 1 0 1 3 s7;3 ¼ 4s5;4 þ 4ð^
vi;j þ 1Þs5;3 þ ð^
vi;j  2w
^i;j Þs5;2  ð2w
^i;j þ 5Þs5;1
3 90 82 81 82 90  5s5;0 ;
1 10 2 1 2 10
0 9 1 0 1 9 s7;4 ¼ 5s5;5 þ ð2^
vi;j þ 5Þs5;4 þ ð2^
vi;j  w
^i;j Þs5;3  ðw
^i;j  4Þs5;2
1 10 2 1 2 10
3 90 82 81 82 90  4s5;1 ;

s7;5 ¼ 3ð^
vi;j þ 2Þs5;5 þ 3^
vi;j s5;4  3ðs5;3 þ s5;2 Þ;

s7;6 ¼ ð4^
vi;j þ w
^i;j Þs5;5 þ ðw
^i;j  2Þs5;4  2s5;3 ;

s7;7 ¼ ð2w
^i;j  1Þs5;5  s5;4 :
The boundary curve S(xi, y) is convex if s7,i > 0, i = 0, 1, 2,
. . . , 7. s7,i > 0, i = 0, 1, 2, . . . , 7 if
( )
Fyi;jþ1  Fyi;j Fyi;jþ1  Fyi;j
v^i;j ¼ w
^i;j > Max 0; ; :
b i;j  Fyi;j Fy  D
D b i;j
i;jþ1

The boundary curve S(xi+1, y) is convex if

( )
Fyiþ1;jþ1  Fyiþ1;j Fyiþ1;jþ1  Fyiþ1;j
v^iþ1;j ¼ w
^iþ1;j > Max 0; ; y :
b iþ1;j  Fy
D F Db iþ1;j
iþ1;j iþ1;jþ1

The above discussion can be wrapped up as:

Algorithm 2. It is the algorithm to apply Theorem 4.

Step 1. Compute ðm þ 1Þ  ðn þ 1Þ convex data points

fðxi ; y j ; F i;j Þ : i ¼ 0; 1; 2; . . . ; m; j ¼ 0; 1; 2; . . . ; ng:
Step 2. Approximate the derivatives F xi;j and F yi;j at knots
Figure 11 Bi-cubic Hermite interpolation. using Section 3.1.
154 M.Z. Hussain et al.

Figure 13 C1 convex rational bi-cubic function. Figure 14 C1 convex rational bi-cubic function.

5. Conclusion

Table 4 A 3D convex data set.

The paper deals with the problem of shape preservation of sur-
y/x 1 2 3 4 5 6 face data. The rational bi-cubic function [6] can proficiently
1 4.1 9.4 23.6 61.8 163.9 438.2 interpolate the data organize over rectangular grids. The mono-
2 9.4 16.9 36.8 87.5 218.1 558.1 tonicity and convexity preservation of data is assured by com-
3 23.6 36.8 69.6 148.4 340.7 819.1 puting constraints on free parameters. All these constraints
4 61.8 87.5 148.4 286.2 603.7 1354.4 involve data and derivatives. The shape preserving interpola-
5 163.9 218.1 340.7 603.7 1177.4 2465.7 tion schemes developed in this paper works well for data with
6 438.2 558.1 819.1 1354.4 2465.7 4843.0
derivative; maintains smoothness and are local. The shape pre-
serving interpolation schemes for convex [5] and monotone [2–
Step 3. Determine the values of the parameters 4] data are futile for data with derivatives. The monotone data
vi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; ^vi;j ; w
^ i;j ; ^viþ1;j ; w
^ iþ1;j by using interpolation scheme [8] does enable derivative specification,
Theorem 4. but fails to maintain smoothness of surface. Moreover, the con-
Step 4. Insert the values of F i;j ; F xi;j F yi;j ; i ¼ 0; 1; 2; vex data interpolation schemes developed in [1,4] are global.
. . . ; m; j ¼ 0; 1; 2; . . . ; n vi;j ; wi;j ; vi;jþ1 ; wi;jþ1 ; ^vi;j ; w
^ i;j ; References
^viþ1;j ; w ^ iþ1;j i ¼ 0; 1; 2; . . . ; m; j ¼ 0; 1; 2; . . . ; n in
rational bi-cubic function (2). [1] Asaturyan, S., 1990. Shape preserving surface interpolation
scheme, Ph.D. Thesis, Scotland (UK): Department of Mathematics
and Computer Science, University of Dundee; 1990.
[2] Beatson RK, Ziegler Z. Monotonicity preserving surface interpo-
4.2.1. Demonstration
lation. SIAM J Numer Anal 1985;22(2):401–11.
[3] Carlson RE, Fritsch FN. Monotone piecewise bicubic interpola-
Example 3. The convex data in Table 3 is computed from the tion. SIAM J Numer Anal 1985;22(2):386–400.
[4] Costantini P, Fontanella F. Shape preserving bivariate interpola-
convex function F(x, y) = x4 + y2. The domain is restricted to
tion. SIAM J Numer Anal 1990;27(2):488–506.
[3, 3] · [3, 3]. [5] Floater MS. A weak condition for the convexity of tensor-product
The convex data in Table 3 is interpolated by bi-cubic Bézier and B-spline surfaces. Adv Comput Math 1994;2(1):67–80.
Hermite (Figs. 11 and 12). It is clear from these figures that bi- [6] Hussain MZ, Sarfraz M, Misbah Irshad, Madeha Amjad. Rational
bi-cubic functions preserving 3D positive data. In: 8th Interna-
cubic Hermite loses convexity for x e [1, 1]. Hence bi-cubic
tional conference, computer graphics, imaging and visualization,
Hermite is unable to preserve the convex shape of data. The Singapore; 16–19 August, 2011. p. 47–52.
convex surface in Fig. 13 is computed by interpolating the [7] Hussain MZ, Sarfraz M, Shaikh TS. Shape preserving rational cubic
same data by convex interpolation scheme of Section 4.2. spline for positive and convex data. Egyptian Inform J 2011;12(3):
231–6.
Example 4.pThe
ffiffiffiffiffiffiffiffiffi convex data in Table 4 is of convex function [8] Sarfraz M, Butt S, Hussain MZ. Surfaces for the visualization of
2 2
Fðx; yÞ ¼ e x þy : The domain is restricted to [1, 6] · [1, 6]. scientific data preserving monotonicity. In: Goodman TNT, Martin
R, editors. Proceedings of the IMA mathematics for surfaces VII,
The convex surface in Fig. 14 is the interpolation of convex September 2–5, 1997, Dundee, UK; 1997. p. 479–95.
data of Table 4. The interpolation is performed by the [9] Sarfraz M, Hussain MZ. Data visualization using rational spline
convexity preserving scheme of Section 4.2. interpolation. J Comput Appl Math 2006;189:513–25.