Subdivision: From Stationary to Non-stationary scheme.

Post on 25-Feb-2016

68 views 1 download

Tags:

description

Subdivision: From Stationary to Non-stationary scheme. Jungho Yoon Department of Mathematics Ewha W. University. Data Type. Sampling/Reconstruction. How to Sample/Re-sample ? - From Continuous object to a finite point set How to handle the sampled data - PowerPoint PPT Presentation

transcript

Subdivision: From Subdivision: From Stationary to Non-Stationary to Non-stationary scheme.stationary scheme.

Jungho YoonDepartment of Mathematics

Ewha W. University

2006.01.09 KMMCS 동서대학교

Data Type

2006.01.09 KMMCS 동서대학교

Sampling/Reconstruction How to Sample/Re-sample ? - From Continuous object to a finite point

set

How to handle the sampled data - From a finite sampled data to a continuous

representation

Error between the reconstructed shape and the original shape

2006.01.09 KMMCS 동서대학교

Subdivision SchemeSubdivision Scheme A simple local averaging rule to build curves and

surfaces in computer graphics

A progress scheme with naturally built-in Multiresolution Structure

One of the most im portant tool in Wavelet Theory

2006.01.09 KMMCS 동서대학교

Approximation Methods

Polynomial Interpolation Fourier Series Spline Radial Basis Function (Moving) Least Square Subdivision Wavelets

2006.01.09 KMMCS 동서대학교

Example

Consider the function

with the data on

2006.01.09 KMMCS 동서대학교

Polynomial Interpolation

2006.01.09 KMMCS 동서대학교

Shifts of One Basis Function Approximation by shifts of one basis

function :

How to choose ?

2006.01.09 KMMCS 동서대학교

Gaussian Interpolation

Subdivision Scheme

Stationary and Non-stationary

2006.01.09 KMMCS 동서대학교

Chainkin’s Algorithm : Chainkin’s Algorithm : corner cuttingcorner cutting

2006.01.09 KMMCS 동서대학교

Deslauriers-Dubuc AlgorithmDeslauriers-Dubuc Algorithm

2006.01.09 KMMCS 동서대학교

SubdivisionSubdivision Non-stationary Butterfly Scheme

2006.01.09 KMMCS 동서대학교

Subdivision SchemeSubdivision Scheme Types

► Stationary or Nonstationary

► Interpolating or Approximating

► Curve or Surface

► Triangular or Quadrilateral

2006.01.09 KMMCS 동서대학교

Subdivision SchemeSubdivision Scheme Formulation

2006.01.09 KMMCS 동서대학교

Subdivision SchemeSubdivision Scheme Stationary Scheme, i.e.,

Curve scheme (which consists of two rules)

2006.01.09 KMMCS 동서대학교

Subdivision : The Limit Subdivision : The Limit FunctionFunction

: the limit function of the subdivision Let Then is called the basic limit funtio

n. In particular, satisfies the two scale relation

2006.01.09 KMMCS 동서대학교

Basic Limit Function : B-splinesBasic Limit Function : B-splines

B_1 spline Cubic spline

2006.01.09 KMMCS 동서대학교

Basic Limit FunctionBasic Limit Function : DD- : DD-schemescheme

2006.01.09 KMMCS 동서대학교

Basic IssuesBasic Issues

Convergence

Smoothness

Accuracy (Approximation Order)

2006.01.09 KMMCS 동서대학교

BBmm-spline subdivision scheme-spline subdivision scheme

Laurent polynomial :

Smoothness Cm-1 with minimal support.

Approximation order is two for all m.

2006.01.09 KMMCS 동서대학교

Interpolatory SubdivisionInterpolatory Subdivision

The general form

4-point interpolatory scheme :

The Smoothness is C1 in some range of w. The Approximation order is 4 with w=1/16.

2006.01.09 KMMCS 동서대학교

Interpolatory SchemeInterpolatory Scheme

2006.01.09 KMMCS 동서대학교

GoalGoal Construct a new scheme which combines the ad

vantages of the aforementioned schemes, while overcoming their drawbacks. Construct Biorthogonal Wavelets

This large family of Subdivision Schemes includes the DD interpolatory scheme and

B-splines up to degree 4.

2006.01.09 KMMCS 동서대학교

Reprod. Polynomials < LReprod. Polynomials < L

Case 1 : L is Even, i.e., L=2N

2006.01.09 KMMCS 동서대학교

Reprod. Polynomials < LReprod. Polynomials < L Case 2 : L is Odd, i.e., L=2N+1

2006.01.09 KMMCS 동서대학교

Stencils of MasksStencils of Masks

2006.01.09 KMMCS 동서대학교

Quasi-interpolatory subdivisionQuasi-interpolatory subdivision General case

L Mask set Sm.

Range of tension

1 O=[v, 1-v] (* If v=1/4, quad spline) E= [1-v, v]

C1 1/4

2 O=[v, 1-2v, v] (* If v= 1/8, cubic spline) E= [1/2, 1/2]

C2 1/8

3 O=[-1/16,9/16,9/16,-1/16] E= [-v, 4v,1-6v,4v,-v]

C2 0.0208<v<0.0404

4 O=[-v,–77/2048+5v,385/512-10v, 385/1024+10v,-55/512-5v,35/2048+v] E(i)=O(7-i) for i=1:6

C3 -0.0106<v<-0.0012

5 O=[3,–25,150,150,–25,3]/256] E=[-v,6v,–15v,1+20v,-15v,6v,-v]

C3 -0.0084<v<-0.0046

2006.01.09 KMMCS 동서대학교

Quasi-interpolatory subdivisionQuasi-interpolatory subdivision

Comparison

CubicB-spline

4-pts interpolatoryscheme

SL Where L=4 (4-5)-scheme

Support of limit ftn [-2, 2] [-3, 3] [-4, 4]

MaximalSmoothness C2 C1 C3

Approximation

Order2 4 4

2006.01.09 KMMCS 동서대학교

Quasi-interpolatory subdivisionQuasi-interpolatory subdivision Basic limit functions for the case L=4

2006.01.09 KMMCS 동서대학교

ExampleExample

2006.01.09 KMMCS 동서대학교

ExampleExample

2006.01.09 KMMCS 동서대학교

Laurent Polynomial

2006.01.09 KMMCS 동서대학교

Smoothness

2006.01.09 KMMCS 동서대학교

Smoothness : Comparison

2006.01.09 KMMCS 동서대학교

Biorthogonal WaveletsBiorthogonal Wavelets

Let and be dual each other if

The corresponding wavelet functions are constructed by

2006.01.09 KMMCS 동서대학교

Symmetric Biorthogonal WaveletsSymmetric Biorthogonal Wavelets

2006.01.09 KMMCS 동서대학교

Symmetric Biorthogonal WaveletsSymmetric Biorthogonal Wavelets

2006.01.09 KMMCS 동서대학교

Nonstationary SubdivisionNonstationary Subdivision

Varying masks depending on the levels, i.e.,

2006.01.09 KMMCS 동서대학교

AdvantagesAdvantages

Design Flexibility

Higher Accuracy than the Scheme based on Polynomial

2006.01.09 KMMCS 동서대학교

Nonstationary SubdivisionNonstationary Subdivision

Smoothness

Accuracy

Scheme (Quasi-Interpolatory)

Non-Stationary Wavelets

Schemes for Surface

2006.01.09 KMMCS 동서대학교

Current Project Construct a new compactly supported biorthogon

al wavelet systems based on Exponential B-splines

Application to Signal process and Medical Imaging (MRI or CT data) Wavelets on special points such GCL points for Numerical PDE

2006.01.09 KMMCS 동서대학교

Thank You !and

Have a Good Tme in Busan!

2006.01.09 KMMCS 동서대학교

Hope to see you in