Weller, Daniel S
Assistant Professor, Electrical and Computer Engineering
- BS, Electrical and Computer Engineering, Carnegie Mellon University
- SM, Electrical Engineering, Massachusetts Institute of Technology
- PhD, Electrical Engineering, Massachusetts Institute of Technology
- Postdoc, Electrical Engineering, University of Michigan
Rice Hall, 85 Engineer's Way, Room 309
P.O. Box 400743
Charlottesville, VA 22904-4743
Signal and Imaging Processing, Advanced Medical Imaging, Signal Estimation and Reconstruction
Increasing quality and resolution requirements in scientific and medical imaging continue to grow rapidly. Meanwhile, consumers demand smaller, longer-lasting imaging sensors for next generation smartphones and wearable devices, at a lower cost. Both trends drive the development of novel signal and image processing algorithms that are faster, more robust, and yield higher quality reconstructions than ever before. Mathematical models for images and measurements and computationally efficient reconstruction algorithms that can exploit these models, are essential to meet this need. Recent application areas of interest include parallel-receive MRI and head motion correction in functional MRI. Exciting research topics in these areas include sparse, structured, and parametric signal models, automatic tuning/regularization parameter selection, iterative algorithms for inverse problems, and synergies between acquisition and reconstruction in image formation. Although solutions are created to address a specific need, the emphasis of the research is on ideas and techniques that can be generalized to address similar signal processing problems found in a variety of other applications.
Regularized iterative methods, such as those my group develops, are revolutionizing the capabilities of image formation and reconstruction in many applications. Many of these reconstructions depend on one or more parameters that can greatly impact the quality of the reproduced images. Unfortunately, selecting appropriate values for these parameters is often nontrivial and can vary from acquisition to acquisition. Automatic parameter selection techniques promise to simplify this process, but parameter tuning still necessitates running many reconstructions in order to evaluate them and settle on the best one. We are investigating reducing the total computational cost of parameter selection to make it practical, via identifying poor parameter choices early in the reconstruction process and using lower-dimensional surrogate problems explicitly for parameter tuning.