Abstract:

Multigrid computational techniques are known to be amongst the
most efficient methods for the numerical solution of several
classes of problems. Originally developed for partial differential
equations in the 1970's, multigrid (more generally, multi-level)
algorithms are currently employed, in academia and industry,
as efficient solvers for an ever increasing variety of linear
and nonlinear problems involving many variables.

In this talk, the multigrid approach will be introduced for a toy
example, followed by a presentation of a few of our recent
developments and applications for problems in image analysis
and processing, as time allows. Our recent and ongoing research
projects include image denoising, binarization, shape-reconstruction
from photometric stereo with constraints, segmentation of images,
quantization, and two-dimensional phase unwrapping.

The talk will require no prior knowledge on the subject. In fact,
anyone well-acquainted with multigrid methods, including anyone who has
taken
my course on the subject, is well-advised not to attend.

(Includes joint work with R. Kimmel, A. Kenigsberg, Y. Koren, A.
Spira, and G. Dardyk).