DSP seminar

DATE: Tuesday, February 10
TIME: 4 p.m.
LOCATION: Beckman 5602

SPEAKER: Chris Hess

TITLE: Maximum Cross-Entropy Generalized Series Imaging

Many imaging applications require the simultaneous resolution
of small spatial structures and closely-spaced temporal events.
This objective is difficult to achieve with Fourier techniques
such as MRI, where spatial and temporal resolution conventionally
come at the expense of one another.  This limitation arises from
the fact that each image in a dynamic time series is represented
independently.  For many problems of interest, the underlying
spatial structure remains the same from image to image.  As a
result, data acquisition requirements can be substantially
relaxed by decoupling the measurement of "static" and "dynamic"
information.  In this talk, I address the question of how to
effectively synthesize these two types of information in order to
improve the overall imaging efficiency.  In particular, I discuss
how the principle of maximum cross-entropy leads to a solution
which allows dynamic images with time-varying intensity and position
to be efficiently represented in terms of a single "reference" image.
The details of this method are discussed, and results from applying
the technique to dynamic MR imaging are presented.