Robert Lee Morrison, Jr., Ph.D.

University of Illinois at Urbana-Champaign, 2007

Joint work with Prof. Minh N. Do, University of Illinois,

and Prof. David C. Munson, Jr., University of Michigan.

Friday, May 2, 1100 am

1500 EECS

This presentation addresses data-driven image restoration for synthetic aperture radar (SAR) imaging systems. Specifically, the autofocus problem in SAR is considered, where the acquired signal data are corrupted by unknown multiplicative phase errors. As a result, the produced imagery is improperly focused. The autofocus problem arises in many important SAR applications, and is of particular concern in next-generation radar systems operating at X-band or higher frequencies where the defocusing effects can be so devastating that the resulting images are almost completely useless.

A critical constraint in the autofocus problem is provided by its inherent multichannel nature, i.e., the image defocusing model exhibits special structure due to the redundancy provided by multiple signal measurements.  By explicitly exploiting the multichannel structure, novel algorithms are developed offering improved restoration performance. We first present a theoretical study providing more insight into metric-based SAR autofocus techniques. Our analytical results show how metric-based autofocus methods implicitly rely on the multichannel defocusing model of SAR autofocus to form well-focused restorations. Utilizing the multichannel structure of the SAR autofocus problem explicitly, we develop a new noniterative restoration approach termed the MultiChannel Autofocus (MCA) algorithm.  In this approach, the focused image is directly recovered using a linear algebraic formulation. Experimental results using actual and simulated SAR data demonstrate that MCA provides superior performance in comparison with existing autofocus methods.