
Citation: Kai Wang, Guicang Zhang. Curve construction based on quartic Bernstein-like basis[J]. AIMS Mathematics, 2020, 5(5): 5344-5363. doi: 10.3934/math.2020343
[1] | Jin Xie, Xiaoyan Liu, Lei Zhu, Yuqing Ma, Ke Zhang . The C3 parametric eighth-degree interpolation spline function. AIMS Mathematics, 2023, 8(6): 14623-14632. doi: 10.3934/math.2023748 |
[2] | Qi Xie, Yiting Huang . A class of generalized quadratic B-splines with local controlling functions. AIMS Mathematics, 2023, 8(10): 23472-23499. doi: 10.3934/math.20231193 |
[3] | Noor Adilla Harim, Samsul Ariffin Abdul Karim, Mahmod Othman, Azizan Saaban, Abdul Ghaffar, Kottakkaran Sooppy Nisar, Dumitru Baleanu . Positivity preserving interpolation by using rational quartic spline. AIMS Mathematics, 2020, 5(4): 3762-3782. doi: 10.3934/math.2020244 |
[4] | Mei Li, Wanqiang Shen . Integral method from even to odd order for trigonometric B-spline basis. AIMS Mathematics, 2024, 9(12): 36470-36492. doi: 10.3934/math.20241729 |
[5] | Abdul Majeed, Muhammad Abbas, Amna Abdul Sittar, Md Yushalify Misro, Mohsin Kamran . Airplane designing using Quadratic Trigonometric B-spline with shape parameters. AIMS Mathematics, 2021, 6(7): 7669-7683. doi: 10.3934/math.2021445 |
[6] | Wan Anis Farhah Wan Amir, Md Yushalify Misro, Mohd Hafiz Mohd . Flexible functional data smoothing and optimization using beta spline. AIMS Mathematics, 2024, 9(9): 23158-23181. doi: 10.3934/math.20241126 |
[7] | Lishan Fang . Smooth digital terrain modeling in irregular domains using finite element thin plate splines and adaptive refinement. AIMS Mathematics, 2024, 9(11): 30015-30042. doi: 10.3934/math.20241450 |
[8] | Salwa Syazwani Mahzir, Md Yushalify Misro, Kenjiro T. Miura . Preserving monotone or convex data using quintic trigonometric Bézier curves. AIMS Mathematics, 2024, 9(3): 5971-5994. doi: 10.3934/math.2024292 |
[9] | Yan Wu, Chun-Gang Zhu . Generating bicubic B-spline surfaces by a sixth order PDE. AIMS Mathematics, 2021, 6(2): 1677-1694. doi: 10.3934/math.2021099 |
[10] | Xiangbin Qin, Yuanpeng Zhu . Static term structure of interest rate construction with tension interpolation splines. AIMS Mathematics, 2024, 9(1): 240-256. doi: 10.3934/math.2024014 |
The construction of basis functions has always been an important topic in CAGD, especially the quadratic and cubic basis functions have a wide range of applications. In [1], Farin gives a kind of piecewise cubic B-spline curves, which are flexible and used conveniently. However, the shape of given curves are fixed to the related control points. For cubic rational B-spline curves, which possess a weight factor can be used to adjust shape of curves, however, improper use of weighting factors can lead to serious parameterization and even destroy the subsequent curve structure[2].
In order to improve the shape and the extent of adjustment of the curve, plenty of methods are generated. Where the most easily method is to add shape parameters to classical cubic Bézier and B-spline curves, the partly can be seen [3,4]. In recent years, researchers proposed some Bernstein basis possessing tension parameters. Hu et al. [5] constructed the developable Bézier-like surfaces using a new kind of cubic Bernstein-like basis with three local shape parameters. In [6], Cravero presented a C3 interpolation function, the shape of the related curves can be adjusted by tension parameters. In [14], Pelosi et al. proposed the quaternion representation for spatial PH cubics and gave results concerning existence of spatial PH cubic interpolants to given G1 Hermite data. In [17], Ait-Haddou et al. proved that the best constrained degree-reduced approximation is qual to the weighted least squares fit of the Bezier coefficients with factored Hahn weights and introduced parameters coming from the Jacobi norm. In [7], Wang constructed a new basis possessing denominator shape parameters, which is extended to B-spline curve and triangular domain. In [15], González constructed C2 algebraic-trigonometric pythagorean splines with different shape parameters. In order not to increase the computational complexity, some scholars have been studying the method of incorporating shape parameters without increasing the orders in n-order Bernstein basis. In [8], Liu gave a set of n-order Bernstein basis with multiple shape parameters in four cases. They have the same order as n-order Bernstein basis. In [9], Xiang improved this result. In the case of odd and even, respectively, a homogenous expansion polynomial with multiple shape parameters of n-order Bernstein basis is given. In [3], Qin et al. also gave the same expansion scheme of n-order Bernstein basis with multiple parameters. In [12], Hu et al. presented a novel shape-adjustable generalized Bézier curve with multiple shape parameters. And then, they extended the basis to rectangular domain[13]. Generally speaking, the extension method which does not increase degree to introduce parameters in Bernstein basis of degree n has faster calculation efficiency, while the extension method which introduces shape parameters by increasing degree of Bernstein basis has stronger ability of adjusting curve. Both of these methods have advantages, so the variable orders as shape parameters is a good choice. In [10], Costantini constructed a set of variable order basis functions in space {1,t,(1−t)m,tn}, and he studied its B-spline form. Subsequently, Costantini [11] extended this space to the space {1,t,t2,⋯,tn−2,(1−t)p,tq}. In[16], Zhu et al. constructed a quasi Bernstein basis with two shape parameters in space {1,3t2−2t3,(1−t)α,tβ}.
The purpose of this paper is to propose a new quartic Bernstein-like basis possessing two exponential shape parameters which serve as a tension parameters. Based on the new basis, a new cubic B-spline-like basis possessing two global and three local shape parameters is presented. And the related cubic B-spline-like curves have C2∩FC2l+3 continuity at each points and include the classical cubic uniform curves as a special case. The rest of this paper are organized as follows: Section 2 gives the definition and properties of the new quartic Bernstein-like basis possessing two exponential shape parameters. And the definition and the connecting conditions of the related curve are shown. Section 3 proposes the cubic B-spline-like basis. And the related properties of connecting, interpolation and locally adjustment of the cubic B-spline-like curves are discussed. The conclusion can be seen in Section 4.
First, we give the definition of quartic Bernstein-like basis.
Definition 1. For any t∈[0,1], let α,β∈[2,+∞), we call the following five functions
{A0(t)=(1−t)α,A1(t)=αt(1−t)α−1,A2(t)=1−∑i≠2Ai,A3(t)=β(1−t)tβ−1,A4(t)=tβ, | (2.1) |
as quartic Bernstein-like basis with two exponential shape parameters α and β.
Remark 1. When α=β=4, the qqr-basis will degenerate to the classical quartic Bernstein basis.
For the convenience of discussion, we will call the quartic Bernstein-like basis as qqr-basis. Also, we denote the qqr-basis functions as Ai(t;α,β)(i=0,1,⋯,4) or Ai(t;α)(i=0,1) and Ai(t;β)(i=3,4). Next, we will give the following lemma, which is used for the analysis of properties of the basis.
Lemma 1. For any t∈[0,1],λ∈[2,+∞), the expression F(t;λ)=1−3t2+2t3−λt(1−t)λ−1−(1−t)λ has a globally minimum value of 0.
Proof. We can easily verify that F(0,λ)=0,F(1,λ)=0,F(t,2), where t∈[0,1],λ∈[2,+∞). For any fixed t∈(0,1), we have
f(t)=(1+t)ln(1−t)+t1−t<0. |
Actually, f′(t)=−t+2ln(1−t)(1−t)2<0 and g(0)=0. Thus, for t∈(0,1),t∈[2,+∞), we have
δF(t,λ)δλ=−(1−t)λ{[1+(λ−1)t]ln(1−t)+t1−t}≥−(1−t)λ[(1+t)ln(1−t)+t1−t]>0. |
Therefore, for any fixed t∈(0,1), F(t,λ) increases with the increases of λ. From this, for arbitrary fixed t∈(0,1), F(t,λ) has a minimum value of F(t,2). Therefore, this lemma holds.
From the above definition of qqr-basis given in (2.1), we have some important properties as follows.
Theorem 1. The qqr-basis given in (2.1) possesses the following properties:
(1) Non-negativity: Ai(t;α,β)(i=0,1,⋯,4)≥0.
(2) Partition of unity: 4∑i=0Ai(t;α,β)=1.
(3) Symmetry: Ai(t;α,β)=A4−i(t;β,α)(i=0,1).
(4) Linear independence: For any (α,β)∈[2,+∞) and t∈[0,1], the qqr-basis Ai(t;α,β)(i=0,1,⋯,4) are linear independent.
(5) Endpoint properties: For any (α,β)∈[2,+∞), we have the following endpoint properties:
A0(0;α,β)=1,A0(1;α,β)=0,A1(0;α,β)=0,A1(1;α,β)=0,A2(0;α,β)=0,A2(1;α,β)=0,A3(0;α,β)=0,A3(1;α,β)=0,A4(0;α,β)=0,A4(1;α,β)=1,A′0(0;α,β)=−α,A′0(1;α,β)=0A′1(0;α,β)=α,A′1(1;α,β)=0,A′2(0;α,β)=0,A′2(1;α,β)=0,A′3(0;α,β)=0,A′3(1;α,β)=−β,A′4(0;α,β)=0,A′4(1;α,β)=β. |
Also, for any (α,β)∈(2,+∞), we have
A″0(0;α,β)=α2−α,A″0(1;α,β)=0,A″1(0;α,β)=2α(1−α),A″1(1;α,β)=0,A″2(0;α,β)=α2−α,A″2(1;α,β)=β2−β,A″3(0;α,β)=0,A″3(1;α,β)=2β(1−β),A″4(0;α,β)=0,A″4(1;α,β)=β2−β. |
Proof. We will proof (1) and (4). The other cases can be easily verified.
(1) We can easily proof that Ai(t;α,β)≥0,i=0,1,3,4. Then, we only consider the basis A2(t;α,β). For α,β∈[2,+∞) and t∈[0,1], from the lemma 1, we have
A2(t;α,β)=[1−3t2+2t3−(1−t)α−αt(1−t)α−1]+[3t2−2t3−β(1−t)tβ−1−tβ]≥0. |
(4) For α,β∈[2,+∞),λi∈R(i=0,1,2,3,4), we have the following linearly combination
4∑i=0λiAi(t;α,β)=0. | (2.2) |
Next, we will differentiate on the above linear combination, we can obtain
4∑i=0λiA′i(t;α,β)=0, | (2.3) |
4∑i=0λiA″i(t;α,β)=0. | (2.4) |
For t=0, the following linear equations can be obtained from (2.2), (2.3) and (2.4):
{λ0=0α(λ1−λ0)=0,(α2−α)(λ0−2λ1+λ2)=0. |
Therefore, it can conclude λ0=λ1=λ2=0. For t=1, from (2.2) and (2.3), we can get λ3=λ4=0, these imply the theorem.
Figure 1 shows the qqr-basis with different shape parameters.
Definition 2. Given control points Pi(i=0,1,⋯,4) in R2 or R3, for any t∈[0,1],α,β∈[2,+∞), we call the expresiion
Q(t;α,β)=4∑i=0Ai(t;α,β)Pi | (2.5) |
as the quartic Bézier-like curves with two exponential shape parameters α and β, where Ai(t;α,β) are qqr-basis given in (2.1). For convenience, we also denote the quartic Bézier-like curves as the qqr-curves.
Theorem 2. The qqr-curves given in (2.5) have some important properties as follows.
(1)Endpoint property:
Q(0;α,β)=P0,Q(1;α,β)=P4. | (2.6) |
Q′(0;α,β)=α(P1−P0),Q′(1;α,β)=β(P4−P3). | (2.7) |
Q″(0;α,β)=(α2−α)(P0−2P1+P2),Q″(1;α,β)=(β2−β)(P2−2P3+P4). | (2.8) |
(2)Symmetry: The control points and define a same qqr-curve, i.e.
Q(t;α,β;P0,P1,P2,P3,P4)=Q(1−t;β,α;P4,P3,P2,P1,P0). | (2.9) |
(3)Geometric invariance: The definition given in (2.5) will meet the folowing rules:
Q(t;α,β;P0+Q∗,P1+Q∗,P2+Q∗,P3+Q∗,P4+Q∗)=Q(t;α,β;P0,P1,P2,P3,P4)+Q∗,Q(t;α,β;P0×M∗,P1×M∗,P2×M∗,P3×M∗,P4×M∗)=Q(t;α,β;P0,P1,P2,P3,P4)×M∗. |
where Q∗ is an any vector in R2 or R3, and M∗ is an any n×n matrix, n=2 or 3.
(4)Convex hull property: The segments of the qqr-curves must lie inside the control polygon spanned by P0,P1,P2,P3,P4.
For any t∈[0,π/2], we rewrite the expression (2.5) as follows:
Q(t;α,β)=P2+A0(t;α)(P0−P2)+A1(t;α)(P1−P2)+A3(t;β)(P3−P2)+A4(t;β)(P4−P2). | (2.10) |
Then, we can easily deduce that the shape parameters α and β can influence four edge ¯P0P2,¯P1P2 and ¯P2P3,¯P2P4, respectively. We also can predict the behavior of the qqr-curves as follows: The local shape of qqr-curves given in (2.10) can be adjusted by modifing parameters α and β. With the increase of α, the curves will tend to the edge ¯P0P2 and ¯P1P2; In contrary, the curves will tend to the opposite direction of edge ¯P0P2 and ¯P1P2. The parameter β has similar adjustment effect as α concerning edge ¯P3P2 and ¯P4P2 as parameter α. However, the curves will tend to the control points P2 when the shape parameters α and β increase at the same time. Thus, we can conclude that the shape parameters α and β serve as tension parameters. In Figure 2, we give some examples about the adjustment effect of parameters on the qqr-curves given in (2.5). Also, the qqr curves have two shape parameters, which can be used to adjust shape of qqr-curves, thus the qqr-curves shows butter approximation ability than classical quartic Bézier curves. Figure 3 illustrates the approximation comparison between qqr-curves and quartic Bézier curves. And the parameters on qqr-curves are set α=β=5.
It is not difficult to find that a single qqr-curves cannot express too complicated geometric design. Thus, the importance of connecting of proposed curve is obvious. Let two qqr-curves be respectively defined as follows:
R1(t;α1,β1)=4∑i=0PiAi(t;α1,β1) | (2.11) |
and
R2(t;α2,β2)=4∑i=0QiAi(t;α2,β2) | (2.12) |
Obviously, if the condition P4=Q0 holds, then the two curve segments (2.11) and (2.12) will form a C0 curve.
But if we want to obtain a more smooth curve, more constrain conditions are needed. For convenience, we rewrite the expression (2.11) and (2.12) as follows:
R(u)={R1(u−u1h1;α1,β1),u∈[u1,u2],R2(u−u2h2;α2,β2),u∈[u2,u3], | (2.13) |
where u1<u2<u3,hi=ui+1−ui,i=1,2.
Theorem 3. For αi,βi∈[2,+∞),i=1,2, if the target curve R(u) satisfies the following condition:
P4=Q0=β1h2P3+α2h1Q1α2h1+β1h2, | (2.14) |
then, we call R(u) is a C1 continuous curve. And if the target curve R(u) satisfy the condition (2.14) and (2.15) simultaneously, we will call R(u) is a C2 continuous curve.
Q2=P4+2h2β1h1α2(P4−P3)+h22(β21−β1)h21(α22−α2)(P2−2P3+P4). | (2.15) |
Proof. According to the formula (2.13), we have
R(u−2)=P4=Q0=R(u+2), |
R′(u−2)=β1(P4−P3)h1=α2(Q1−Q0)h2=R′(u+2), | (2.16) |
R″(u−2)=(β21−β1)(P2−2P3+P4)h21=(α22−α2)(Q0−2Q1+Q2)h22=R″(u+2). | (2.17) |
Through simple calculations, we can easily conclude this Theorem holds.
It was points out in [18,19] that Frent continuity(FC) is a reasonable smoothness property for practical applications. From the end-point properties (Theorem 2.(1)) of the qqr-curves, we can conclude the following Theorem.
Theorem 4. For αi,βi∈[l+3,+∞), if the formula (2.14) and (2.15) hold simultaneously, the target curve R(u) is a C2∩FC2l+3 continuous curve at the knot u2, where i=1,2 and l∈Z+.
Proof. For k=3,4,⋯,2l+3, it follows that
R(k)(u−2)=[A(k)2(1)P2+A(k)3(1)P3+A(k)4(1)P4]hk1, |
R(k)(u+2)=[A(k)0(0)Q0+A(k)1(0)Q1+A(k)2(0)Q2]hk2. |
From (2.1), we have
A(k)0(0)+A(k)1(0)+A(k)2(0)=0,A(k)2(1)+A(k)3(1)+A(k)4(1)=0. |
And from (2.16) and (2.17), we have
P4−P3=h1R′(u2)β1,P2−P3=h21R″(u2)β21−β1−h1R′(u2)β1,Q1−Q0=h2R′(u2)α2,Q2−Q1=h22R″(u2)α22−α2+h2R′(u2)α2. |
Thus, we can conclude
R(k)(u−2)=[A(k)2(1)(P2−P3)+A(k)4(1)(P4−P3)]hk1=1hk1{A(k)2(1)h21R″(u2)β21−β1+h1[A(k)4(1)−A(k)2(1)]R′(u2)β1}, |
R(k)(u+2)=[A(k)0(0)(Q0−Q1)+A(k)2(0)(Q2−Q1)]hk2=1hk2{A(k)2(0)h22R″(u2)α22−α2+h2[A(k)2(0)−A(k)1(0)]R′(u2)α2}. |
By directly computing, it follows that
R(k)(u+2)−R(k)(u−2)=[A(k)2(0)hk−22(α22−α2)−A(k)2(1)hk−21(β21−β1)]R″(u2)+[A(k)2(0)−A(k)0(0)hk−12α2−A(k)4(1)−A(k)2(1)hk−11β1]R′(u2). |
These imply the theorem.
Figure 4. gives some examples of continuous curves with different conditions at the knot u2. For C1 continuous curve, the related parametric values are set α1=10,β1=3,α2=3,β=5. For C2 continuous curve, the related parametric values are set α1=10,β1=2,α2=2,β=6. For C2∩FC3 continuous curve, the related parametric values are set α1=10,β1=4,α2=4,β=5.5. For C2∩FC5 continuous curve, the related parametric values are set α1=10,β1=5.5,α2=5.5,β=5.5.
In this subsection, based on the proposed qqr-basis given in (2.1), we will construct a set of cubic B-spline-like basis.
Given a any set of knots u0<u1<⋯<un+4, we refer to U=(u0,u1,⋯,un+4) as a knot vector. Let hi=ui+1−ui,t∈[0,1], for any i∈Z+,α,β∈[2,+∞),xi,yi,zi∈R, we have
μi=hihi+hi+1,υi=hi+1hi+hi+1, |
ρi=μ2i+1(1−zi+1)+υ2i+1zi+2μi+1υi+1(1−xi+1−yi+1),Bi,0(t)=μixiA3(t;β)+μi(μizi+2υixi)A4(t;β),Bi,1(t)=μi(μizi+2υixi)A0(t;α)+(μizi+υixi)A1(t;α)+ziA2(t;α,β)+[μi+1(1−xi+1−yi+1)+υi+1zi]A3(t;β)+ρiA4(t;β),Bi,2(t)=ρiA0(t;α)+[μi+1(1−zi+1)+υi+1(1−xi+1−yi+1)]A1(t;α)+(1−zi+1)A2(t;α,β)+[μi+2yi+2+υi+2(1−zi+1)]A3(t;β)+υi+2[2μi+2yi+2+υi+2(1−zi+1)]A4(t;β),Bi,3(t)=υi+2[2μi+2yi+2+υi+2(1−zi+1)]A0(t;α)+υi+2yi+2A1(t;α), |
where Aj(t;α,β)(j=0,1,⋯,4) are the qqr-basis functions given in (2.1).
Definition 3. Given a knot vector U=(u0,u1,⋯,un+4), let tj(u)=(u−uj)/hj, where j=0,1,⋯,n+3, the related cubic B-spline-like basis with two global shape parameters α,β and three local shape parameters xi,yi,zi are defined as follows
Bi(u)={Bi,0(ti),u∈[ui,ui+1),Bi,1(ti+1),u∈[ui+1,ui+2),Bi,1(ti+2),u∈[ui+2,ui+3),Bi,1(ti+3),u∈[ui+3,ui+4),0,u∉[ui+4,ui+5), | (3.1) |
where i=0,1,⋯,n.
For convenience, we redefine the cubic B-spline-like basis as qqr B-spline basis. We will also call Bi(u) as a uniform basis when given a set of equidistant knots, and the related knot vector U is called as a uniform knot vector. Similarly, for the case of non-equidistant knots, Bi(u) and U are called as non-uniform basis and knot vector, respectively.
Remark 2. When the global shape parameters are set α=β=4 and all xi,yi,zi satisfy the following rules, then the qqr B-spline basis will degenerate to cubic non-uniform B-spline basis.
xi=hi4(hi+hi+1+hi+2),yi=hi+14(hi−1+hi+hi+1),zi=2hi+hi+12(hi+hi+1+hi+2). |
Figure 5. shows the qqr B-spline basis with different global and local shape parameters.
Theorem 5. For any u∈[u3,un+1], the qqr B-spline basis functions satisfy the following rule:
n∑i=0Bi(u)=1. | (3.2) |
Proof. For any u∈[ui,ui+1),i=3,4,⋯,n, we have Bj(u)=0,j≠i−3,i−2,i−1,i. Since
Bi−3(u)=Bi−3,3(ti),Bi−2(u)=Bi−2,2(ti),Bi−1(u)=Bi−1,1(ti),Bi(u)=Bi−3,0(ti). |
It follows that
n∑j=0Bj(u)=i∑j=i−3Bj(u)=4∑j=0Aj(ti;α,β)=1. |
These imply the theorem.
Theorem 6. For any α,β∈[2,+∞) and ui<u<ui+4, if xi,yi,zi satisfy the following rules, then we have Bi(u)>0.
{xi≥0,yi+2≥0,xi+1+bi+1≤1,zi≥0,zi+1≤1. | (3.3) |
Proof. We can easily find that it is direct application of (3.1).
Theorem 7. For any i∈Z+,α,β∈(2,+∞), if xi,yi,zi satisfy the following form:
−yi−1zi−1+(zi−1−xi−1)(1−zi−2)≠0, | (3.4) |
then, the system {B0(u),B1(u),⋯,Bn(u)} is linear independent on [u3,un+1].
Proof. For any ai∈R(i=0,1,⋯,n),u∈[u3,un+1], we let
M(u)=2∑i=0aiBi(u)=0. | (3.5) |
For α,β∈(2,+∞), by directly computing, we have
M(ui)=υi−1[2μi−1yi−1+υi−1(1−zi−2)]ai−3+ρi−2ai−2+μi−1(μi−1zi−1+2υi−1xi−1)ai−1=0,M′(ui)=αhi−1+hi{[(2μi−1−1)yi−1+υi−1(1−zi−2)](ai−2−ai−3)+[(1−2μi−1)xi−1+μi−1zi−1](ai−1−ai−2)}=0,M″(ui)=α2−α(hi−1+hi)2[(2yi−1+zi−2−1)(ai−2−ai−3)+(zi−1−2xi−1)(ai−1−ai−2)]=0, |
where i=3,4,⋯,n+1. Thus, the linear equations with respect to ai−3,ai−2,ai−1 are defined as follows:
{υi−1[2μi−1yi−1+υi−1(1−zi−2)]ai−3+ρi−2ai−2+μi−1(μi−1zi−1+2υi−1xi−1)ai−1=0,[(2μi−1−1)yi−1+υi−1(1−zi−2)](ai−2−ai−3)+[(1−2μi−1)xi−1+μi−1zi−1](ai−1−ai−2)=0,(2yi−1+zi−2−1)(ai−2−ai−3)+(zi−1−2xi−1)(ai−1−ai−2)=0. |
Thus, we can easily get the coefficient matrix from the above linear equations. By directly computing, we have υi−1[2μi−1yi−1+υi−1(1−zi−2)]+ρi−2+μi−1(μi−1zi−1+2υi−1xi−1)=1. Therefore, the coefficient matrix can be rewrite as follows:
|Di|=|υi−1[2μi−1yi−1+υi−1(1−zi−2)]1μi−1(μi−1zi−1+2υi−1xi−1)−[(2μi−1−1)yi−1+υi−1(1−zi−2)]0(1−2μi−1)xi−1+μi−1zi−1−(2yi−1+zi−2−1)0zi−1−2xi−1|=−yi−1zi−1+(zi−1−xi−1)(1−zi−2). |
From the formula (3.4), we can get |Di|≠0. Thus, we have ai−3=ai−2=ai−1=0(i=3,4,⋯,n+1). These imply the theorem.
Theorem 8. For any α=β∈(2,+∞), the qqr B-spline basis Bi(u) is C2 continuity.
Proof. For any α=β∈(2,+∞), from Theorem 1(5), we have
B(k)i(u+i)=B(k)i(u+i+4)=0(k=0,1). |
By directly computing, it follows that
Bi(u±i+1)=μi(μizi+2υixi),B′i(u±i+1)=α(μizi+2υixi−xi)hi+hi+1,Bi(u±i+2)=ρi,B′i(u±i+2)=α[μi+1(xi+1+yi+1−zi+1)+υi+1(1−zi−xi+1−yi+1)]hi+1+hi+2,Bi(u±i+3)=υi+2[2μi+2yi+2+υi+2(1−zi+1)],B′i(u±i+3)=α[(1−2μi+2)yi+2−υi+2(1−zi+1)]hi+2+hi+3. |
Moreover, we have
B″i(u+i)=B″i(u−i+4)=0,B″i(u±i+1)=(α2−α)(zi−2xi)(hi+hi+1)2,B″i(u±i+2)=(α2−α)(2xi+1+2yi+1+zi−zi+1−1)(hi+1+hi+2)2,B″i(u±i+3)=(α2−α)(1−zi+1−2yi+2)(hi+2+hi+3)2. |
These imply the theorem.
So far, we have been discussing the situation of a single knot, and the conclusion is relatively simple. However, it is relatively important for the case where the knots multiplicity λ≤4. Thus, for the case of multiple knots, from Definition 3, we can redefine it as follows:
Bi(u)={Bi,1(ti+1),u∈[ui+1,ui+2),Bi,2(ti+2),u∈[ui+2,ui+3),Bi,3(ti+3),u∈[ui+3,ui+4),0,u∉[ui,ui+4). |
where ui=ui+1 is a double knot.
We give the following theorem, useful for the geometric discussion of the case of multiple knots.
Theorem 9. For any fixed u, we suppose that the qqr B-spline basis has a λ multiplicity knot, then the continuity of the qqr B-spline basis at point u is reduced from Cj−1 to Cj−λ, where λ=2,3,j=2, for α=β=2 and λ=2,3,4,j=3, for α=β∈(2+∞). In addition, the support interval of the qqr B-spline basis is reduced from 4 segments to 5−λ segments.
Proof. That is a direct application of (3.1) and the related expression shown in the previous sections.
Definition 4. Given control points Pi(i=0,1,⋯,n)∈R2orR3 and a knot vector U=(u0,u1,⋯,un+4), then for n≥3,u∈[u3,un+1], we call
Q(u)=n∑i=0Bi(u)Pi | (3.6) |
as a cubic B-spline-like curve with two global shape parameters α,β∈[2,+∞) and three local shape parameters xi,yi,zi, where Bi(u) is the qqr B-spline basis given in (3.1).
For convenience, we redefine the cubic B-spline-like curve as qqr B-spline curve. Moreover, for u∈[ui,ui+1],3≤i≤n, the qqr B-spline curve segments can be defined as follows:
Qi(u)=i∑j=i−3Bj(u)Pj. | (3.7) |
Theorem 10. The qqr B-spline curves have the following properties:
(1)When the qqr B-spline basis satisfies the conditions of Remark 2, then the qqr B-spline curve will degenerate to cubic B-spline curve.
(2)For u∈[ui,ui+1], if the Theorem 5 and 6 hold simultaneously, the qqr B-spline curves will lie in the convex hull of the control points Pi−3,Pi−2,Pi−1,Pi.
(3)Given a non-uniform knots vector, if ui is a vector which has multiplicity λ, then the qqr B-spline curves are Cj−λ continuous at ui, where λ=1,2,3,j=2, for α=β=2, and λ=1,2,⋯,4, for α=β∈(2,+∞). Also, for i=3,4,⋯,n+1, if ui is a single knot, we can get
Q(ui)=υi−1[2μi−1yi−1+υi−1(1−zi−2)]Pi−3+ρi−2Pi−2+μi−1(μi−1zi−1+2υi−1xi−1)Pi−1,Q′(ui)=αhi−1+hi{[(2μi−1−1)yi−1+υi−1(1−zi−2)](Pi−2−Pi−3)+[(1−2μi−1)xi−1+μi−1zi−1](Pi−1−Pi−2)},Q″(ui)=(α2−α)(hi−1+hi)2[(2yi−1+zi−1−1)(Pi−2−Pi−3)+(zi−1−2xi−1)(Pi−1−Pi−2)]. |
Proof. We can conclude that the Theorem holds from the result of Theorem 8 and 9. These imply the Theorem.
Based on the qqr B-spline curves, we give the curve segment of the Bézier form, which is useful understanding of role of the shape parameters.
For convenience, we let
Ri−1,i=(1−zi)Pi−1+ziPi,Si−1=yiPi−2+(1−xi−yi)Pi−1+aiPi,Ti−1=υ2iRi−2,i−1+2υiμiSi−1+μ2iRi−1,i, |
where Ti is adjustable interpolation points. Thus, we can rewrite the qqr B-spline segments given in (3.7) as
Q(u)=A0(ti;α,β)Ti−2+A1(ti;α,β)(υi−1Si−2+μi−1Ri−2,i−1)+A2(ti;α,β)Ri−2,i−1+A3(ti;α,β)(υiRi−2,i−1+μiSi−1)+A3(ti;α,β)Ti−1. | (3.8) |
From the above preparation, we have the following discussion.
We give two knots ui and ui+1(ui<ui+1), where ui is a quadruple point when i=3 and ui+1 is a quadruple point when i=n, thus, we can get hi−2=hi−1=hi+1=hi+2. Also, let xi−1=0,yi=0,zi−2=0,zi=1, for u∈[ui,ui+1], we can get the following basis functions:
{Bi−3(u)=A0(ti;α,β)+yi−1A1(ti;α,β),Bi−2(u)=(1−yi−1)A1(ti;α,β)+(1−zi−1)A2(ti;α,β),Bi−1(u)=zi−1A2(ti;α,β)+(1−xi)A3(ti;α,β),Bi(u)=xiA3(ti;α,β)+A4(ti;α,β). | (3.9) |
Remark 3. Within the framework of qqr B-spline segments given in (3.7), the basis functions given in (3.9) degenerate to the classical cubic Bernstein basis when the global parameters are set α=β=4 and the local parameters are set xi=yi−1=1/4,zi−1=1/2. Thus, the related curves are the classical cubic Bézier curves.
Theorem 11. The basis functions given in (3.9) have the following properties:
(1)For any u∈[ui,ui+1],i∑j=i−3Bj(u)=1.
(2)For any u∈[ui,ui+1],Bj(u)>0(j=i−3,i−2,⋯,i) when 0≤xi,yi−1,ci−1≤1.
(3)Together with (3.7) and (3.9), we can easily find the following interesting phenomenon. With the decrease of the parameters xi and yi−1, the related curve will moves to the edge ¯Pi−2Pi−1 of the control polygon. With the decrease of the parameters zi−1, the curve will moves to the point Pi−2, on the contrary, the curve will moves to the point Pi−1.
(4)Together with (3.7) and (3.9), we have
Q(ui)=Pi−3,Q(ui+1)=Pi, |
Q′(ui)=α(1−yi−1)hi(Pi−2−Pi−3),Q′(ui+1)=α(1−xi)hi(Pi−Pi−1), | (3.10) |
Q″(ui)=(α2−α)h2i[zi−1(Pi−1−Pi−2)−(1−2yi−1)(Pi−2−Pi−3)], | (3.11) |
Q″(ui)=β2−βh2i[(1−2xi)(Pi−Pi−1)−(1−zi−1)(Pi−1−Pi−2)]. | (3.12) |
Proof. The theorem can be verified from previous discussion.
Figure 6. gives some examples of the Bézier form curves of qqr B-spline curves. The first row shows the influence of local shape parameters on curves, where the global shape parameters are fixed, and we set all α=β=4. The second row shows the influence of global shape parameters on curves, where the local shape parameters are fixed, and we set all xi=yi=0.2,zi=0.5.
First, we consider the curve of C2 continuity. Then the following Theorem can be concluded.
Theorem 12. For any α=β∈(2,+∞), the qqr B-spline curves are C2 continuity.
Proof. From Theorem. 8 and 10(3), it is easily conclude that this Theorem holds.
Next, we consider the curve of C2∩FC2l+3, we have the following theorem.
Theorem 13. Let xi−1=yi−1=0,zi−2≠1 and zi−1≠0, for α=β∈[l+3,+∞), the qqr B-spline curves Q(u) are C2∩FC2l+3 continuous at point ui, where i=3,4,⋯,n and l∈Z+.
Proof. From (3.11) and (3.12), we have
Pi−1−Pi−2=1zi−1[hi−1+hiαQ′(ui)+hi(hi−1+hi)α2−αQ″(ui)], | (3.13) |
Pi−2−Pi−3=11−zi−2[hi−1+hiαQ′(ui)−hi−1(hi−1+hi)α2−αQ″(ui)]. | (3.14) |
For j=3,4,⋯,2l+3, by directly computing, we have
Q(j)(u+i)−Q(j)(u−i)=[B(j)i−3,3(0)hji−B(j)i−3,2(1)hji−1]Pi−3+[B(j)i−2,2(0)hji−B(j)i−2,1(1)hji−1]Pi−2+[B(j)i−1,1(0)hji−B(j)i−1,0(1)hji−1]Pi−1. | (3.15) |
From the Theorem 1(5), we can get
[B(j)i−3,3(0)hji−B(j)i−3,2(1)hji−1]+[B(j)i−2,2(0)hji−B(j)i−2,1(1)hji−1]+[B(j)i−1,1(0)hji−B(j)i−1,0(1)hji−1]=0hji−0hji−1=0. |
Therefore, we have
Q(j)(u+i)−Q(j)(u−i)=[−B(j)i−3,3(0)hji+B(j)i−3,2(1)hji−1](Pi−2−Pi−3)+[B(j)i−1,1(0)hji−B(j)i−1,0(1)hji−1](Pi−1−Pi−2). |
From (3.13) and (3.14), we have
Q(j)(u+i)−Q(j)(u−i)={hi(hi−1+hi)(α2−α)zi−1[B(j)i−1,1(0)hji−B(j)i−1,0(1)hji−1]−hi−1(hi−1+hi)(α2−α)(1−zi−1)[−B(j)i−3,3(0)hji+B(j)i−3,2(1)hji−1]}Q″(ui)+{hi−1+hiαzi−1[B(j)i−1,1(0)hji−B(j)i−1,0(1)hji−1]+hi−1+hiα(1−zi−2)[−B(j)i−3,3(0)hji+B(j)i−3,2(1)hji−1]}Q′(ui) |
These imply the theorem.
Let α=β, for C2∩FC2l+3(k∈Z+) continuity, with regard to the curve segments given in (3.7), it follows that
Bi−3(u)=υ2i−1(1−zi−2)A0(ti;α,α),Bi−2(u)=ρi−2A0(ti;α,α)+(1−μi−1zi−1)A0(ti;α,α)+(1−zi−1)A2(ti;α,α)+υi(1−zi−1)A3(ti;α,α)+υ2i(1−zi−1)A4(ti;α,α),Bi−1(u)=μ2i−1zi−1A0(ti;α,α)+μi−1zi−1A1(ti;α,α)+zi−1A2(ti;α,α)+(μi+υizi−1)A3(ti;α,α)+ρi−1A4(ti;α,α),Bi(u)=μ2iziA4(ti;α,α). |
By directly computing, we have
hiαQ′(u)=υ2i−1(1−zi−2)(1−ti)α−1(Pi−2−Pi−3)+{υi−1zi−1[μi−1(1−ti)+(α−1)ti](1−ti)α−2+μi(1−zi−1)[(α−1)(1−ti)+υiti]tα−2i}(Pi−1−Pi−2)+μ2izitα−1i(Pi−Pi−1). | (3.16) |
From the above formula given in (3.16), we can easily find that Q′(u) is a non-negative linear combination of the {Pj+1−Pj}. Further, for C2∩FC2l+3(l∈Z+) continuity, we have
h2iα(α−1)Q″(u)=−υ2i−1(1−ti)α−2(1−zi−2)(Pi−2−Pi−3)+μ2itα−2izi(Pi−Pi−1)+(α+1){υi−1zi−1(1−t)α−3[(1−μi−1)(1−ti)−(α−2)ti]+μi(1−zi−1)tα−3i[(υi−1)ti+(α−2)(1−t)]}(Pi−1−Pi−2). | (3.17) |
In this case, we suppose Q″(u) is a non-negative linear combination concerning the form {(Pj+1−Pj)−φ(Pj−Pj−1)}, where ϕ,φ≥0. Moreover, from zi−1(μi−1−1)=0 and μi(1−zi−1)=0, we will get hi=0. Thus, we refer to the Q(u) as convexity preserving.
Figure 7. shows the planar and space C2∩FC5 continuous curves (Left) and cubic B-spline curves (Right). All parameters are set α=β=4,xi=yi=0,zi=0.5. From the figure we can find that C2∩FC5 continuous curves are nearer the control polygon than the cubic B-spline curves.
In this subsection, we will use the qqr B-spline curves given in (3.6) to interpolate locally, the resulting curves will have C2 continuity. First, we let
{xi−1=−hi−1hizi−1,yi−1=−hi2hi−1(1−zi−2). | (3.18) |
For i=3,4,⋯,n+1, from the Theorem 10(3), we have
Q(ui)=Pi−2, | (3.19) |
Q′(ui)=α2(hi−1+hi)[hi(1−zi−2)Pi−2−Pi−3hi−1+hi−1zi−1Pi−1−Pi−2hi], | (3.20) |
Q″(ui)=α2−αhi−1+hi[zi−1Pi−1−Pi−2hi−(1−zi−2)Pi−2−Pi−3hi−1]. | (3.21) |
In order to obtain a open curve, we will add two additional segments, that is P−1=P0,Pn+1=Pn. Therefore, by using (3.18), the interpolate points P0 and Pn need to add i=2 and i=n+2.
For any i, with the conditions (3.18), the resulting curves can be C2 continuity by interpolating all given points. And the resulting curves have two global shape parameters and a local shape parameter. At knot ui, when α=4, based on the formula (3.20), we have
Q′(ui)=2hi−1+hi[hi(1−zi−2)Pi−2−Pi−3hi−1+hi−1zi−1Pi−1−Pi−2hi], |
In this case, when zi−2=zi−1=0.5, the interpolation curve has the following tangent
Q′(ui)=1hi−1+hi[hiPi−2−Pi−3hi−1+hi−1Pi−1−Pi−2hi], |
which can generate the Hermite interpolation curves.
Figure 8. gives some examples of the interpolation qqr B-spline curves with different shape parameters. In the left figure, we show the interpolation qqr B-spline curves with different shape parameters, the global shape parameters are set α=β=4. In the right figure, we show the interpolation qqr B-spline curves with different shape parameters, all the local shape parameters are set zi=0.5.
The qqr B-spline curve have two global shape parameters α,β and three local shape parameters xi,yi,zi. From the Theorem 13, we can find that the global parameters α,β can effectively generate C2∩FC2l+3 continuous curves and can global modify the shape of the curve. When we fix the global shape parameters α,β, we also locally adjust the shape of the qqr B-spline curves by modifying the local shape parameters xi,yi,zi.
From the formula (3.7), the curve segments Q(u) can be adjusted by modifying the local shape parameters xi−1,xi,yi−1,yi,zi−2,zi−1,zi when u∈[ui,ui+1]. Thus, we can easily get that the different parameters xi,yi only affect two segments which defined in the interval u∈[ui,ui+2], but the parameters zi can affect three segments which defined in the interval u∈[ui,ui+3].
From the formula (3.7) and Theorem 10(3), the shape of the proposed qqr B-spline curves can be easily predicted. First, we fix the global shape parameters. Then, the related coefficient values of Pi−2 and Pi−1 increase and coefficient values of Pi−3 and Pi decrease with the decrease of the parameter xi and yi, respectively. With regard to the parameters yi−1 and xi−1, the curve have the same adjustment effect with the coefficient values of Pi−3,Pi−1 and Pi−2,Pi, respectively. Thus, we can find the parameters xi and yi have tension effect. With decrease of the parameter xi and yi, the curve Q(u) will move to the edge ¯Pi−2Pi−1 of the control polygon. On the contrary, the curve will move to the opposite direction. Moreover, the shape parameters zi serves as local control bias. With the decrease of zi−1, the end points Q(ui) and Q(ui+1) of the qqr B-spline curve segments will move to the control points Pi−2. In the opposite case, the curve segments will move to the control points Pi−1.
Figure 9. gives some examples of the adjustment effect of the qqr B-spline curves. In the left figure, we show the effect of local shape parameters on qqr B-spline curves, the global shape parameters are set α=β=4. In the right figure, we show the effect of global shape parameters on qqr B-spline curves, all the local shape parameters are set xi=yi=0.01,zi=0.5.
A new quartic Bernstein-like basis possessing two exponential shape parameters is given, which is useful for CAGD. And the conditions of C2∩FC2l+3 continuity and definition of the related curves are discussed. Based on the new basis, a new cubic B-spline-like basis possessing two global and three local shape parameters is presented. And the related cubic B-spline-like curves have C2∩FC2l+3 continuity at each points and include the classical cubic uniform curves as a special case. The related properties of connecting, interpolation and local adjustment of the cubic B-spline-like curves are discussed. How to accurately quantify the influence of each parameter on the shape of the curve will be our future work.
This research was supported by National Natural Science Foundation of China (61861040), the Gansu education Department Science and Technology Achievement Transformation project (no. 2017D-09) and Gansu Province Science, technology project funding (no.17YF1FA119) and Lanzhou Science and Technology Plan Project (2018-4-35).
The authors declare that they have no conflicts of interest.
[1] | G. Farin, Curves and Surfaces for Computer Aided Geometric Design: A Practical Guide, Academic Press, Inc., 1993. |
[2] | L. Piegl, W. Tiller, The NURBS Book, New York, Springer, 1995. |
[3] | X. Q. Qin, G. Hu, N. Zhang, et al. A novel extension to the polynomial basis functions describing Bezier curves and surfaces of degree n with multiple shape parameters, Appl. Math. Comput., 223 (2013), 1-16. |
[4] |
X. A. Han, Y. C. Ma, X. L. Huang, A novel generalization of Bezier curves and surface, J. Comput. Appl. Math., 217 (2008), 180-193. doi: 10.1016/j.cam.2007.06.027
![]() |
[5] |
G. Hu, H. X. Cao, S. Zhang, et al. Developable Bézier-like surfaces with multiple shape parameters and its continuity conditions, Appl. Math. Model., 45 (2017), 728-747. doi: 10.1016/j.apm.2017.01.043
![]() |
[6] |
I. Cravero, C. Manni, Shape-preserving interpolants with high smoothness, J. Comput. Appl. Math., 157 (2003), 383-405. doi: 10.1016/S0377-0427(03)00418-7
![]() |
[7] | K. Wang, G. C. Zhang, New trigonometric basis possessing denominator shape parameters, Math. probl. Eng., 2018 (2018), 1-25. |
[8] | Z. Liu, Representation and approximation of curves and surfaces based on the Bezier method in CAGD, Hefei University of technology, 2009. |
[9] | T. Xiang, Z. Liu, W. Wang, et al. A novel extension of Bézier curves and surfaces of the same degree, J. Inf. Comput. Sci., 7 (2010), 2080-2089. |
[10] |
P. Costantini, Curve and surface construction using variable degree polynomial splines, Comput. Aided Geom. D., 17 (2000), 419-466. doi: 10.1016/S0167-8396(00)00010-8
![]() |
[11] |
P. Costantini, T. Lyche, C. Manni, On a class of weak Tchebysheff systems, Numer. Math., 101 (2005), 333-354. doi: 10.1007/s00211-005-0613-6
![]() |
[12] |
G. Hu, J. L. Wu, X. Q. Qin, A novel extension of the Bézier model and its applications to surface modeling, Adv. Eng. Softw., 125 (2018), 27-54. doi: 10.1016/j.advengsoft.2018.09.002
![]() |
[13] | G. Hu, C. C. Bo, G. Wei, et al. Shape-adjustable generalized Bézier surfaces: Construction and its geometric continuity conditions, Appl. Math. Comput., 378 (2020), 125215. |
[14] |
F. Pelosi, R. T. Farouki, C. Manni, et al. Geometric Hermite interpolation by spatial Pythagoreanhodograph cubics, Adv. Comput. Math., 22 (2005), 325-352. doi: 10.1007/s10444-003-2599-x
![]() |
[15] |
C. González, G. Albrecht, M. Paluszny, et al. Design of algebraic-trigonometric pythagorean hodograph splines with shape parameters, Comput. Appl. Math., 37 (2018), 1472-1495. doi: 10.1007/s40314-016-0404-y
![]() |
[16] |
Y. Zhu, X. Han, S. Liu, Curve construction based on four αβ-Bernstein-like basis functions, J. Comput. Appl. Math., 273 (2015), 160-181. doi: 10.1016/j.cam.2014.06.014
![]() |
[17] |
R. Ait-Haddou, M. Bartoň, Constrained multi-degree reduction with respect to Jacobi norms, Comput. Aided Geom. D., 42 (2016), 23-30. doi: 10.1016/j.cagd.2015.12.003
![]() |
[18] | J. Hoschek, D. Lasser, Fundamentals of Computer Aided Geometric Design, A. K. Peters, Ltd, 1993. |
[19] |
P. Constantini, C. Manni, Geometric construction of spline curves with tension properties, Comput. Aided Geom. D., 20 (2003), 579-599. doi: 10.1016/j.cagd.2003.06.009
![]() |