Shaping Regularization - PowerPoint PPT Presentation

1 / 20
About This Presentation
Title:

Shaping Regularization

Description:

Shaping regularization is fast convergent. Compared to the regularization operator, shaping operator is often easier to design, ... – PowerPoint PPT presentation

Number of Views:152
Avg rating:3.0/5.0
Slides: 21
Provided by: Yax
Category:

less

Transcript and Presenter's Notes

Title: Shaping Regularization


1
Shaping Regularization
Best SEG Talk Sergey Fomels
  • Yaxun Tang

2
Paper Name
  • Shaping regularization in geophysical estimation
    problems
  • Author Sergey Fomel
  • SI 2.4 Wave-Equation and Topology

3
Introduction
well-posed
Model
Indirect constraints
Geophysical Estimation
Insufficient Data
ill-posed
4
Introduction
  • Traveltime tomography
  • Migration velocity analysis
  • High-resolution Radon transform
  • Spectral decomposition

5
Regularization theory
  • Goal of inversion
  • Fit the observed data
  • Fit the estimated model to a priorly assumed
    behavior (Tikhonov)
  • i.e. to minimize vector

L forward operator m model d data D
regularization operator scaling parameter
6
Regularization theory
  • The formal solution
  • We can also minimize the least-square norm of the
    vector

P model preconditioning operator
mPp
7
Regularization Theory
  • The formal solution
  • Those two solutions are equivalent, if

8
Smoothing
  • Find a model m that fits the observed data d but
    is in a certain sense smoother
  • In this case LI

9
Regularized smoothing
10
Triangle smoothing
11
Shaping regularization
  • Take smoothing as a fundamental operation!
  • Smoothing implies projection of the input model
    to the space of admissible functions
  • Shaping just translates the input into an
    acceptable model

projection operator shaping
12
Shaping regularization
  • Define
  • Then, we get

13
Shaping regularization
  • Introducing a scaling of L
  • When

Suitable for an iterative inversion of the
conjugate-gradient method
14
1-D data regularization
15
1-D data regularization
Regularization
Model Preconditioning
16
1-D data regularization
S is lowpass filtering with a Gaussian
smoother
S is bandpass filtering with a shifted Gaussian
17
1-D data regularization
18
Velocity estimation
19
(No Transcript)
20
Conclusion
  • The main idea of shaping regularization is to
    take shaping as a fundamental operation
  • Shaping regularization is fast convergent
  • Compared to the regularization operator, shaping
    operator is often easier to design,
Write a Comment
User Comments (0)
About PowerShow.com