The goal of this research is to develop models, estimation techniques, and computational algorithms for inverse problems arising in geophysics, and in particular in reservoir characterization. Although large volumes of data may be available in these problems, they do not sufficiently determine the underground structure under study. We are studying the use of stochastic models and nonlinear constraints to decrease this uncertainty.

Our goal is to develop optimal techniques and efficient computational algorithms in three areas of imaging: (1) image reconstruction from partial information; (2) acquisition of time-varying images; (3) visualization of vector fields. We are studying multiresolution and nonlinear techniques for tomographic with limited angle data, and for other ill-posed inverse problems. We are also developing a systematic theory for designing minimum rate sampling patterns. We are developing algorithms for maximally informative display of vector-valued images, such as are acquired in multispectral or multimodality remote sensing and diagnostic imaging. This research has applications in biomedical imaging; video; remote sensing and surveillance; and geophysics.

This research studies the basic issues in using sensor arrays to image dynamic scenes, including the detection, estimation, and correction of motion, the optimal design of image acquisition systems, and visualization of complex time-varying multidimensional data. We focus on the case of significant object motion during the acquisition of one image frame, with specific emphasis on synthetic aperture radar (SAR) and time-varying vector fields (TVVF). We are analyzing the effects of scene motion imagery in both spotlight mode SAR and inverse SAR, and developing algorithms for producing sharply focused images, with velocity estimates for all moving targets.

Fluid flow data are four-dimensional (x,y,z,t) and vector-valued (velocity, pressure, mass density). Processing such immense data sets is a great challenge. We are studying three aspects: (1) data compression, (2) interpolation, and (3) detection, recognition, and tracking of interesting patterns especially in turbulent flow. Since turbulent flowfields contain similar coherent patterns at many different scales, fractals and related techniques may prove very useful.

We use the term HCI in a very broad sense to include communication between person and computer as well as communication between persons via computer. An example of the former is a person using a workstation, an example of the latter is tele-collaboration. We are investigating a variety of issues related to the use of computer vision in HCI. These include: facial feature extraction and tracking, determining 3-D head pose, facial movement modeling, analysis, and synthesis, hand gesture recognition, human body motion analysis, and person identification.

We are developing algorithms for segmenting still and moving images using multiple cues including color and motion. We are concentrating on images of outdoor scenes in the context of applications to autonomous navigation.

The motion of three types of nonrigid objects is studied: left ventricle of heart, human head/face, and fluid. The goal is to explore and develop concepts, methodologies, and techniques that will be useful for a wide class of non rigid objects. Special attention is paid to motion modeling.

This research aims at generating new technologies for nondestructive evaluation of damage of large structures such as bridges and domes. The structure is loaded dynamically, and the locations and movements of point and edge features on the structure are measured by CCD cameras. The damage is inferred from the measured data.

This project investigates how the learning process in adaptive digital filters is disturbed by hardware failures and how to design filters and adaptive algorithms that can continue operating in the presence of such failures. Because adaptive systems are capable of adjusting system parameters to reduce a specified error criterion, it has been shown that whenever a hardware failure occurs that increases the error, the system will attempt to compensate for this failure by further self-adjustment. Recently this research concentrated on FFT-based fault-tolerant filter structures and new, efficient data-reusing algorithms to achieve fault-tolerant behavior under a broad class of hardware failures.

This research considers combining system-level modular redundancy with the arithmetic modularity of residue number system (RNS) arithmetic to achieve fault tolerance in high-speed DSP systems. Double, triple, and quadruple modular redundancy, system-level concepts which are frequently used in commercial fault-tolerant computers are combined with RNS modularity for realizing important DSP computational kernels. As a potential application, a block LMS (BLMS) adaptive algorithm has been proposed for high-speed VLSI adaptive filters. A chip design for an adaptive filter based on a serial-by-modulus RNS architecture algorithm and an RNS implementation with double modular redundancy is being developed, and it is expected that experimental chips will be fabricated by MOSIS to provide proof of conceptions.

This research project has the broad objective of developing adaptive techniques for processing multidimensional signals and signal representations. Many multidimensional signals of interest exhibit nonstationary or time-varying behavior that can be dealt with effectively only by adapting the processing to the current signal characteristics. Images and sensor array data are two types of multidimensional signals of great military and commercial interest. This research attempts to develop ``dual-use'' technology (military and commerical) through the use of time-frequency signal analysis, two-dimensional adaptive signal processing, and real-time video signal processing to achieve new ways of processing multidimensional information.

The goal of this research project is to develop methods based on time-frequency analysis for estimating the relative energies of individual components of complex signals. Time-frequency techniques are potentially attractive for this application because of their ability to separate signal components that significantly overlap in time or frequency. Linear, quadratic, and adaptive time-frequency analysis and synthesis methods will be investigated.

We are investigating synthetic aperture radar (SAR) as a means of imaging runways through fog and cloud cover from an approaching aircraft. Current radars with traditional signal processing are incapable of providing the resolution required at long ranges, because of the wide beam widths of the antennas employed. Our approach uses the changes in angular aspect of points on the runway, provided by the motion of the aircraft, to produce high-resolution imagery from return signals collected by a conventional radar.

* Denotes principal investigator.