CS 790 - Math 752: Inverse Problems in Imaging -
Spring 2010
Instructor
Lectures
The class meets Tuesdays and Thursdays from 9:30 to 10:45 in
Manchester, Room 124.
Prerequisites
Calculus of Several Variables, Linear Algebra, and a beginning course in
Numerical Methods (or, you should review this material). Matlab will be used in the course.
Text - No formal text is required. Material from the following
books will be used.
"Computational Methods for Inverse Problems", by C. Vogel, SIAM Press, Philadelphia, 2002.
"Deblurring Images: Matrices, Spectra, and Filtering", P. Hansen, J. Nagy, and D. O'Leary,
SIAM Press, Philadelphia, 2006. Paperback copies are available. We will also use
other resources, and some lecture notes will be provided.
Course Themes
Inverse problems are problems where causes for a desired or an observed
effect are to be determined. They have, nearly always driven by
applications, been studied for nearly a century now. An important key
feature, both theoretically and numerically, of inverse problems is
their ill-posedness, i.e., they do not fulfill Hadamard's classical
requirements of existence, uniqueness and stability, under data
perturbations, of a solution: Solutions of an inverse problem might not
exist for all data (e.g., a consistent temperature history exists only
for a very smooth final temperature in the model of the classical heat
equation), it might not be unique (which raises the practically
relevant question of identifiability, i.e., the question if the data
contain enough information to determine the desired quantity), and it
might be unstable with respect to data perturbations. The last aspect
is of course especially important, since in real-world problems,
measurements always contain noise (another source of noise being errors
in numerical procedures), and approximation methods for solving inverse
problems which are as insensitive to noise as possible have to be
constructed, so-called regularization methods.
The course will concern "Emerging Applications of Inverse Problems
Techniques to Imaging Science". These include medical
imaging, atmospheric imaging, biometrics, integrating
optics and imaging, and computer vision techniques.
We will concentrate on: (1) mathematical principles, including
variational PDE methods, useful in solving
inverse
problems in science, medicine, and engineering,
(2) techniques for solving modest size inverse problems using the
Matlab Image Processing and Optimization Toolboxes, (3)
methods for solving large scale
inverse problems.
Problem Sets & Grading
There will be 3 projects (1 individual/2 group), a midterm
(take home), and a final exam (take home). Course grades will be determined
using the following weighting: individual problem set - 10%, group
problem sets with class presentations - 20% each, mid-term and final
exams - 25% each.