Tensor Product Spline Surfaces

Spline in the x-direction:
s 1 = ∑ i = 1 n 1 c i ϕ i ( x ) s_{1} = \sum_{i=1}^{n_{1}}c_{i}\phi_{i}(x) s1=i=1n1ciϕi(x)

Let c i c_{i} ci be the spline in the y-direction:
s 2 = c i ( y ) = ∑ j = 1 n 2 c i , j φ j ( y ) s_{2} = c_{i}(y) = \sum_{j=1}^{n_{2}}c_{i,j}\varphi_{j}(y) s2=ci(y)=j=1n2ci,jφj(y)

Then, combine s 1 s_{1} s1 and s 2 s_{2} s2, we get the function of a Tensor Product Spline Surface:
f ( x , y ) = ∑ i = 1 n 1 ∑ j = 1 n 2 c i , j ϕ i ( x ) φ j ( y ) f(x,y) = \sum_{i=1}^{n_{1}}\sum_{j=1}^{n_{2}}c_{i,j}\phi_{i}(x)\varphi_{j}(y) f(x,y)=i=1n1j=1n2ci,jϕi(x)φj(y)

where the coefficients ( c i , j ) i , j = 1 n 1 , n 2 (c_{i,j})_{i,j=1}^{n_{1},n_{2}} (ci,j)i,j=1n1,n2 can be real numbers. This linear space of functions is denoted s 1 ⊗ s 2 s_{1} \otimes s_{2} s1s2.

The space s 1 ⊗ s 2 s_{1} \otimes s_{2} s1s2 is spanned by the functions ϕ i ( x ) φ j ( y ) i , j = 1 n 1 , n 2 {\phi_{i}(x)\varphi_{j}(y)}_{i,j=1}^{n_{1},n_{2}} ϕi(x)φj(y)i,j=1n1,n2 and therefore has dimension n 1 n 2 n_{1}n_{2} n1n2.

Tensor product surface in the matrix form:
f ( X , Y ) = ϕ ( X ) T C φ ( Y ) f(X,Y) = \phi(X)^{T}C\varphi(Y) f(X,Y)=ϕ(X)TCφ(Y)

where ϕ = ( ϕ 1 , ϕ 2 , … , ϕ n 1 ) T \phi = (\phi_{1}, \phi_{2}, \dots, \phi_{n_{1}})^{T} ϕ=(ϕ1,ϕ2,,ϕn1)T, φ = ( φ 1 , φ 2 , … , φ n 2 ) T \varphi = (\varphi_{1}, \varphi_{2}, \dots, \varphi_{n_2})^{T} φ=(φ1,φ2,,φn2)T, and C = ( c i , j ) C = (c_{i,j}) C=(ci,j) is the matrix of coefficients.

Advantages over other ways constructing surfaces:

Many standard operations that we wish to perform with the surfaces are very simple generalisations of corresponding univariate operations.

The approximation methods that we developed for functions and curves can be utilized directly from approximation for surfaces.

knots with multiplicity

  • Each knot of multiplicity k k k reduces at most k − 1 k-1 k1 basis functions’ non-zero domain.
  • At each internal knot of multiplicity k k k, the number of non-zero basis functions is at most p − k + 1 p-k+1 pk+1, where p p p is the degree of the basis functions.
Tensor Product Spline Interplotion

∑ p = 1 m 1 ∑ q = 1 m 2 c p , q ϕ q ( y j ) φ p ( x i ) = g ( x i , y j ) = f i , j \sum_{p=1}^{m_{1}}\sum_{q=1}^{m_{2}}c_{p,q}\phi_{q}(y_{j})\varphi_{p}(x_{i}) = g(x_{i},y_{j}) = f_{i,j} p=1m1q=1m2cp,qϕq(yj)φp(xi)=g(xi,yj)=fi,j

This double sum can be split into two sets of simple sums:
∑ p = 1 m 1 d p , j ϕ p ( x i ) = f i , j ∑ q = 1 m 2 c p , q φ q ( y j ) = d p , j \begin{aligned} \sum_{p=1}^{m_{1}}d_{p,j}\phi_{p}(x_{i}) &= f_{i,j} \\ \sum_{q=1}^{m_{2}}c_{p,q}\varphi_{q}(y_{j}) &= d_{p,j} \end{aligned} p=1m1dp,jϕp(xi)q=1m2cp,qφq(yj)=fi,j=dp,j

We define the matrices:
Φ = ( ϕ i , p ) ∈ ℜ m 1 , m 2 , ϕ i , p = ϕ p ( x i ) , Ψ = ( φ j , q ) ∈ ℜ m 1 , m 2 , φ j , p = φ q ( y j ) , D = ( d p , j ) ∈ ℜ m 1 , m 2 , F = ( f i , j ) ∈ ℜ m 1 , m 2 C = ( c p , q ) ∈ ℜ m 1 , m 2 \begin{aligned} \Phi &= (\phi_{i,p}) \in \Re^{m_{1},m_{2}}, \hspace{1cm} \phi_{i,p} = \phi_{p}(x_{i}), \\ \Psi &= (\varphi_{j,q}) \in \Re^{m_{1},m_{2}}, \hspace{1cm} \varphi_{j,p} = \varphi_{q}(y_{j}), \\ D &= (d_{p,j}) \in \Re^{m_{1},m_{2}}, \hspace{1cm} F = (f_{i,j}) \in \Re^{m_{1},m_{2}} \\ C &= (c_{p,q}) \in \Re^{m_{1},m_{2}} \end{aligned} ΦΨDC=(ϕi,p)m1,m2,ϕi,p=ϕp(xi),=(φj,q)m1,m2,φj,p=φq(yj),=(dp,j)m1,m2,F=(fi,j)m1,m2=(cp,q)m1,m2

then
∑ p = 1 m 1 d p , j ϕ p ( x i ) = ∑ p = 1 m 1 ϕ i , p d p , j = ( Φ D ) i j = ( F ) i j ∑ q = 1 m 2 c p , q φ q ( y j ) = ∑ q = 1 m 2 φ j , p c p , q = ( Ψ C T ) j p = ( D T ) j p \begin{aligned} \sum_{p=1}^{m_{1}} d_{p,j}\phi_{p}(x_{i}) &= \sum_{p=1}^{m_{1}}\phi_{i,p}d_{p,j} = (\Phi D)_{ij} = (F)_{ij} \\ \sum_{q=1}^{m_{2}} c_{p,q}\varphi_{q}(y_{j}) &= \sum_{q=1}^{m_{2}}\varphi_{j,p}c_{p,q} = (\Psi C^{T})_{jp} = (D^{T})_{jp} \end{aligned} p=1m1dp,jϕp(xi)q=1m2cp,qφq(yj)=p=1m1ϕi,pdp,j=(ΦD)ij=(F)ij=q=1m2φj,pcp,q=(ΨCT)jp=(DT)jp

Thus:
Φ D = F a n d C Ψ T = D \Phi D = F \hspace{1cm} and \hspace{1cm} C\Psi^{T} = D ΦD=FandCΨT=D

If the matrices Φ \Phi Φ and Ψ \Psi Ψ are nonsigular, there is a unique g ∈ s 1 ⊗ s 2 g \in s_{1} \otimes s_{2} gs1s2 such that g ( x i , y j ) = f i , j g(x_{i},y_{j}) = f_{i,j} g(xi,yj)=fi,j, for i ∈ { 1 , … , m 1 } i \in \{1,\dots,m_{1}\} i{1,,m1}, j ∈ { 1 , … , m 2 } j \in \{1,\dots,m_{2}\} j{1,,m2}.

Least Square for Gridded Data

Given data
( x i , y j , f i , j ) i = 1 , j = 1 m 1 , m 2 (x_{i},y_{j},f_{i,j})_{i=1,j=1}^{m_{1},m_{2}} (xi,yj,fi,j)i=1,j=1m1,m2

positive weights ( w i ) i = 1 m 1 (w_{i})_{i=1}^{m_{1}} (wi)i=1m1 and ( v j ) j = 1 m 2 (v_{j})_{j=1}^{m_{2}} (vj)j=1m2, and univariate spline spaces s 1 s_{1} s1 and s 2 s_{2} s2, find a spline surface g g g in s 1 ⊗ s 2 s_{1} \otimes s_{2} s1s2 which solves the minimization problem
min ⁡ g ∈ s 1 ⊗ s 2 ∑ i = 1 m 1 ∑ j = 1 m 2 w i v j [ g ( x i , y j ) − f i , j ] 2 \min_{g \in s_{1} \otimes s_{2}}\sum_{i=1}^{m_{1}}\sum_{j=1}^{m_{2}}w_{i}v_{j}[g(x_{i},y_{j})-f_{i,j}]^2 gs1s2mini=1m1j=1m2wivj[g(xi,yj)fi,j]2

Its matrix form
min ⁡ C ∈ ℜ n 1 , n 2 ∥ A C B T − G ∥ 2 \min_{C \in \Re^{n_{1},n_{2}}} \parallel ACB^{T}-G \parallel^{2} Cn1,n2minACBTG2

where
A = ( a i , p ) ∈ ℜ m 1 , n 1 a i , p = w i ϕ p ( x i ) B = ( b j , q ) ∈ ℜ m 2 , n 2 b j , q = v j φ q ( y j ) G = ( w i v j f i , j ) ∈ ℜ m 1 , m 2 C = ( c p , q ) ∈ ℜ n 1 , n 2 \begin{aligned} A &= (a_{i,p}) \in \Re^{m_{1},n_{1}} \hspace{1cm} a_{i,p} = \sqrt{w_{i}}\phi_{p}(x_{i}) \\ B &= (b_{j,q}) \in \Re^{m_{2},n_{2}} \hspace{1cm} b_{j,q} = \sqrt{v_{j}}\varphi_{q}(y_{j}) \\ G &= (\sqrt{w_{i}}\sqrt{v_{j}}f_{i,j}) \in \Re^{m_{1},m_{2}} \\ C &= (c_{p,q}) \in \Re^{n_{1},n_{2}} \end{aligned} ABGC=(ai,p)m1,n1ai,p=wi ϕp(xi)=(bj,q)m2,n2bj,q=vj φq(yj)=(wi vj fi,j)m1,m2=(cp,q)n1,n2

Here, the norm ∥ ⋅ ∥ 2 \parallel \cdot \parallel^{2} 2 is the Frobenius norm,
∥ E ∥ = ( ∑ i = 1 m ∑ j = 1 n ∣ e i , j ∣ 2 ) 1 2 \parallel E \parallel = (\sum_{i=1}^{m}\sum_{j=1}^{n}\vert e_{i,j}\vert^{2})^{\frac{1}{2}} E=(i=1mj=1nei,j2)21

for any rectangular m × n m \times n m×n matrix E = ( e i , j ) E = (e_{i,j}) E=(ei,j).

min ⁡ C ∈ ℜ n 1 , n 2 ∥ A C B T − G ∥ 2 \min_{C \in \Re^{n_{1},n_{2}}} \parallel ACB^{T} -G \parallel^{2} minCn1,n2ACBTG2 always has a solution C = C ∗ C=C^{*} C=C, and the solution is unique iif both matrices A A A and B B B have linearly independent columns. The solution C ∗ C^{*} C can be found by solving the matrix equation
A T A C ∗ B T B = A T G B A^{T}AC^{*}B^{T}B = A^{T}GB ATACBTB=ATGB

Parametric Surfaces

A function of two variables E = h ( x , y ) E = h(x,y) E=h(x,y) can always be considered as a parametric surface through the representation f ( u , v ) = ( u , v , h ( u , v ) ) f(u,v) = (u,v,h(u,v)) f(u,v)=(u,v,h(u,v)).

A parametric representation of class C m C^{m} Cm of a set ς ⊆ ℜ 3 \varsigma \subseteq\Re^{3} ς3 is a mapping f f f of an open set Ω ⊆ ℜ 2 \Omega \subseteq \Re^{2} Ω2 onto ς \varsigma ς such that

(1) f f f has continous derivatives up to order m m m,

(2) the Jacbian Matrix of f f f given by
J ( f ) = ( D u f 1 ( u , v ) D v f 1 ( u , v ) D u f 2 ( u , v ) D v f 2 ( u , v ) D u f 3 ( u , v ) D v f 3 ( u , v ) ) J(f) = \begin{pmatrix} D_{u}f^{1}(u,v) & D_{v}f^{1}(u,v) \\ D_{u}f^{2}(u,v) & D_{v}f^{2}(u,v) \\ D_{u}f^{3}(u,v) & D_{v}f^{3}(u,v) \\ \end{pmatrix} J(f)=Duf1(u,v)Duf2(u,v)Duf3(u,v)Dvf1(u,v)Dvf2(u,v)Dvf3(u,v)

has full rank k k k for all ( u , v ) (u,v) (u,v) is Ω \Omega Ω.

The unit normal of the regular parametric representation f f f is the vector
N ( u , v ) = D u f ( u , v ) × D v f ( u , v ) ∥ D u f ( u , v ) × D v f ( u , v ) ∥ N(u,v) = \frac{D_{u}f(u,v) \times D_{v}f(u,v)}{\parallel D_{u}f(u,v) \times D_{v}f(u,v) \parallel} N(u,v)=Duf(u,v)×Dvf(u,v)Duf(u,v)×Dvf(u,v)

The normal vector plays an important role when we start analysing the curvature of surfaces.

Let ς \varsigma ς be a surface with a regular parametric representation f f f. The tangent space or tangent plane T f ( u , v ) T_{f(u,v)} Tf(u,v) of ς \varsigma ς at f ( u , v ) f(u,v) f(u,v) is the plane in R 3 \R^{3} R3 spanned by the two vectors D u f ( u , v ) D_{u}f(u,v) Duf(u,v) and D v f ( u , v ) D_{v}f(u,v) Dvf(u,v).i.e., all vectors on the form δ 1 D u f ( u , v ) + δ 2 D v f ( u , v ) \delta_{1}D_{u}f(u,v) + \delta_{2}D_{v}f(u,v) δ1Duf(u,v)+δ2Dvf(u,v). The normal of the tangent plane T f ( u , v ) T_{f(u,v)} Tf(u,v) is the normal vector N ( u , v ) N(u,v) N(u,v).

Parametric Tensor Product Spline Surfaces

A parametric tensor product spline surface is given by a parametric representation on the form
f ( u , v ) = ∑ i = 1 m ∑ j = 1 n c i j B i , d , δ ( u ) B j , l , τ ( v ) f(u,v) = \sum_{i=1}^{m}\sum_{j=1}^{n}c_{ij}B_{i,d,\delta}(u)B_{j,l,\tau}(v) f(u,v)=i=1mj=1ncijBi,d,δ(u)Bj,l,τ(v)

where the coefficients ( c i j ) i , j = 1 m , n (c_{ij})_{i,j=1}^{m,n} (cij)i,j=1m,n are point in space,
c i j = ( c i j 1 , c i j 2 , c i j 3 ) c_{ij} = (c_{ij}^{1},c_{ij}^{2},c_{ij}^3) cij=(cij1,cij2,cij3)
and δ = ( δ i ) i = 1 m + d + 1 \delta = (\delta_{i})_{i=1}^{m+d+1} δ=(δi)i=1m+d+1 and τ = ( τ j ) j = 1 n + l + 1 \tau = (\tau_{j})_{j=1}^{n+l+1} τ=(τj)j=1n+l+1 are knot vectors for splines of degree d d d and l l l.

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值