| |
| |
Foreword | |
| |
| |
Preface | |
| |
| |
| |
Introduction | |
| |
| |
| |
An Illustrative Example | |
| |
| |
| |
Regularization by Filtering | |
| |
| |
| |
A Deterministic Error Analysis | |
| |
| |
| |
Rates of Convergence | |
| |
| |
| |
A Posteriori Regularization Parameter Selection | |
| |
| |
| |
Variational Regularization Methods | |
| |
| |
| |
Iterative Regularization Methods | |
| |
| |
Exercises | |
| |
| |
| |
Analytical Tools | |
| |
| |
| |
Ill-Posedness and Regularization | |
| |
| |
| |
Compact Operators, Singular Systems, and the SVD | |
| |
| |
| |
Least Squares Solutions and the Pseudo-Inverse | |
| |
| |
| |
Regularization Theory | |
| |
| |
| |
Optimization Theory | |
| |
| |
| |
Generalized Tikhonov Regularization | |
| |
| |
| |
Penalty Functionals | |
| |
| |
| |
Data Discrepancy Functionals | |
| |
| |
| |
Some Analysis | |
| |
| |
Exercises | |
| |
| |
| |
Numerical Optimization Tools | |
| |
| |
| |
The Steepest Descent Method | |
| |
| |
| |
The Conjugate Gradient Method | |
| |
| |
| |
Preconditioning | |
| |
| |
| |
Nonlinear CG Method | |
| |
| |
| |
Newton's Method | |
| |
| |
| |
Trust Region Globalization of Newton's Method | |
| |
| |
| |
The BFGS Method | |
| |
| |
| |
Inexact Line Search | |
| |
| |
Exercises | |
| |
| |
| |
Statistical Estimation Theory | |
| |
| |
| |
Preliminary Definitions and Notation | |
| |
| |
| |
Maximum Likelihood Estimation | |
| |
| |
| |
Bayesian Estimation | |
| |
| |
| |
Linear Least Squares Estimation | |
| |
| |
| |
Best Linear Unbiased Estimation | |
| |
| |
| |
Minimum Variance Linear Estimation | |
| |
| |
| |
The EM Algorithm | |
| |
| |
| |
An Illustrative Example | |
| |
| |
Exercises | |
| |
| |
| |
Image Deblurring | |
| |
| |
| |
A Mathematical Model for Image Blurring | |
| |
| |
| |
A Two-Dimensional Test Problem | |
| |
| |
| |
Computational Methods for Toeplitz Systems | |
| |
| |
| |
Discrete Fourier Transform and Convolution | |
| |
| |
| |
The FFT Algorithm | |
| |
| |
| |
Toeplitz and Circulant Matrices | |
| |
| |
| |
Best Circulant Approximation | |
| |
| |
| |
Block Toeplitz and Block Circulant Matrices | |
| |
| |
| |
Fourier-Based Deblurring Methods | |
| |
| |
| |
Direct Fourier Inversion | |
| |
| |
| |
CG for Block Toeplitz Systems | |
| |
| |
| |
Block Circulant Preconditioners | |
| |
| |
| |
A Comparison of Block Circulant Preconditioners | |
| |
| |
| |
Multilevel Techniques | |
| |
| |
Exercises | |
| |
| |
| |
Parameter Identification | |
| |
| |
| |
An Abstract Framework | |
| |
| |
| |
Gradient Computations | |
| |
| |
| |
Adjoint, or Costate, Methods | |
| |
| |
| |
Hessian Computations | |
| |
| |
| |
Gauss--Newton Hessian Approximation | |
| |
| |
| |
A One-Dimensional Example | |
| |
| |
| |
A Convergence Result | |
| |
| |
Exercises | |
| |
| |
| |
Regularization Parameter Selection Methods | |
| |
| |
| |
The Unbiased Predictive Risk Estimator Method | |
| |
| |
| |
Implementation of the UPRE Method | |
| |
| |
| |
Randomized Trace Estimation | |
| |
| |
| |
A Numerical Illustration of Trace Estimation | |
| |
| |
| |
Nonlinear Variants of UPRE | |
| |
| |
| |
Generalized Cross Validation | |
| |
| |
| |
A Numerical Comparison of UPRE and GCV | |
| |
| |
| |
The Discrepancy Principle | |
| |
| |
| |
Implementation of the Discrepancy Principle | |
| |
| |
| |
The L-Curve Method | |
| |
| |
| |
A Numerical Illustration of the L-Curve Method | |
| |
| |
| |
Other Regularization Parameter Selection Methods | |
| |
| |
| |
Analysis of Regularization Parameter Selection Methods | |
| |
| |
| |
Model Assumptions and Preliminary Results | |
| |
| |
| |
Estimation and Predictive Errors for TSVD | |
| |
| |
| |
Estimation and Predictive Errors for Tikhonov Regularization | |
| |
| |
| |
Analysis of the Discrepancy Principle | |
| |
| |
| |
Analysis of GCV | |
| |
| |
| |
Analysis of the L-Curve Method | |
| |
| |
| |
A Comparison of Methods | |
| |
| |
Exercises | |
| |
| |
| |
Total Variation Regularization | |
| |
| |
| |
Motivation | |
| |
| |
| |
Numerical Methods for Total Variation | |
| |
| |
| |
A One-Dimensional Discretization | |
| |
| |
| |
A Two-Dimensional Discretization | |
| |
| |
| |
Steepest Descent and Newton's Method for Total Variation | |
| |
| |
| |
Lagged Diffusivity Fixed Point Iteration | |
| |
| |
| |
A Primal-Dual Newton Method | |
| |
| |
| |
Other Methods | |
| |
| |
| |
Numerical Comparisons | |
| |
| |
| |
Results for a One-Dimensional Test Problem | |
| |
| |
| |
Two-Dimensional Test Results | |
| |
| |
| |
Mathematical Analysis of Total Variation | |
| |
| |
| |
Approximations to the TV Functional | |
| |
| |
Exercises | |
| |
| |
| |
Nonnegativity Constraints | |
| |
| |
| |
An Illustrative Example | |
| |
| |
| |
Theory of Constrained Optimization | |
| |
| |
| |
Nonnegativity Constraints | |
| |
| |
| |
Numerical Methods for Nonnegatively Constrained Minimization | |
| |
| |
| |
The Gradient Projection Method | |
| |
| |
| |
A Projected Newton Method | |
| |
| |
| |
A Gradient Projection-Reduced Newton Method | |
| |
| |
| |
A Gradient Projection-CG Method | |
| |
| |
| |
Other Methods | |
| |
| |
| |
Numerical Test Results | |
| |
| |
| |
Results for One-Dimensional Test Problems | |
| |
| |
| |
Results for a Two-Dimensional Test Problem | |
| |
| |
| |
Iterative Nonnegative Regularization Methods | |
| |
| |
| |
Richardson--Lucy Iteration | |
| |
| |
| |
A Modified Steepest Descent Algorithm | |
| |
| |
Exercises | |
| |
| |
Bibliography | |