Research Goals

The Computational Imaging Group (CIG) at Washington University in St. Louis pursues research on the development of advanced algorithms and mathematical tools for imaging. Topics of interest are image reconstruction, optimization, machine learning, and statistical inference. Research efforts are taking place at two complementary levels:

  • Fundamental and mathematical aspects of imaging
  • Application-oriented projects in biomedical or industrial imaging

Some of our recent research is summarized below.

Learning for Imaging

Best computational imaging methods must exploit intricate relationships and features in the data. However, finding ways to design such methods in practice is difficult. Our goal is to leverage the latest progress in machine learning and nonconvex optimization in order to develop algorithms capable of delivering highest quality results in shortest amount of time. Our recent contributions include MMSE-ISTA and Online Multimodal Convolutional Learning.

Imaging Through Scattering

Much of modern biomedical imaging is based on linear models that assume that photons travel following a straight path. This makes corresponding imaging methods inaccurate for many applications by placing fundamental limits—in terms of resolution, penetration, and quality. Our goal is to leverage advanced optimization tools in order to enable imaging under nonlinear scattering scenarios. Our research into this fascinating direction has produced advanced imaging methods such as Learning Tomography and SEAGLE.

Compressive Imaging

The number of measurements provided by an imaging instrument is often limited due to hardware constraints. Compressive imaging is when the total number of measurements falls below the number of pixels/voxels in the image. This makes image reconstruction an underdetermined problem with fewer measurements than unknowns. Our goal is to develop methods that enable successful recovery of the image from compressive measurements by designing algorithms that maximally exploit the redundancies in the signal. Recent contributions include Parallel-Proximal TV and Motion-Adaptive Regularization.

Collaborations

  • Dr. Emrah Bostan, UC Berkeley, Berkeley, USA
  • Dr. Petros Boufounos, MERL, Cambridge, USA
  • Prof. Alyson Fletcher, UCLA, Los Angeles, USA
  • Prof. Vivek Goyal, Boston University, Boston, USA
  • Dr. Dehong Liu, MERL, Cambridge, USA
  • Prof. Laurent Jacques, UCL, Louvain-la-Neuve, Belgium
  • Dr. Hassan Mansour, MERL, Cambridge, USA
  • Prof. Demetri Psaltis, EPFL, Lausanne, Switzerland
  • Prof. Sundeep Rangan, New York University, New York, USA
  • Prof. Anne Sentenac, Institut Fresnel, Marseille, France
  • Prof. Philip Schniter, Ohio State University, Columbus, USA
  • Prof. Lei Tian, Boston University, Boston, USA
  • Prof. Michael Unser, EPFL, Lausanne, Switzerland
  • Dr. Anthony Vetro, MERL, Cambridge, USA
  • Prof. Laura Waller, UC Berkeley, Berkeley, USA
  • Dr. Brendt Wohlberg, Los Alamos National Laboratory, Los Alamos, USA