Visual data plays key roles in numerous domains, ranging from medical imaging, microscopy, astronomy, security and autonomous vehicles, to personal photography and social media. Dr. Michaeli’s research addresses theoretical and algorithmic challenges relating to the acquisition, restoration, enhancement, manipulation, and editing of visual data. The last decade has seen tremendous progress in image processing methods, particularly with the surge of deep learning techniques. However, their widespread adoption, for example in scientific domains, is still hindered by fundamental limitations, like small training data-sets, inherent ambiguities in the imaging processes, and lack of proper ways to incorporate physical knowledge into deep learning based methods.
Dr. Michaeli strives to provide theoretical analyses of the inherent limits and trade-offs that govern imaging problems, and to develop new algorithms that will lay the basis for the next generation of imaging systems.