NRAO/Socorro Colloquium Series:
17 November 1995
Richard Puetter
Univ. of California at San Diego
Information, Language, Bayesian Estimation,
and
Pixon-Based Image Restoration
Pixon-based image reconstruction is a new and highly successful
method of performing Bayesian image reconstruction. Compared to other
methods, e.g. Maximum Likelihood (ML) and Maximum Entropy (ME),
pixon-based methods provide enhanced resolution (factor of a few),
often much greater (more than an order of magnitude) sensitivity to
faint sources, and robust rejection of spurious sources. These gains
are obtained by characterizing the information content in the data
using a more natural language (pixons) for representing an image than
provided by competing techniques. The advances of this approach can be
understood from first-principle information theory, and the direct
consequences of this is to provide a highly optimized Bayesian prior.
Indeed, pixon-based reconstruction can be shown to be a "super-theory"
of ML and ME image reconstruction, i.e. the pixon-based solution is the
best solution optimized over a very large solution space containing
both the ML and ME solution spaces. Numerous examples of pixon-based
image reconstruction from IR, X-ray, and gamma-ray astronomy will be
shown.
Friday, 17 November 1995
11:00am
Array Operations Center Auditorium
Local Host: Tim Cornwell
Other NRAO/Socorro colloquia
mrupen@nrao.edu