Structured Low-Rank Matrix Recovery via Optimization Methods
Abstract: Motivated by applications from single-molecule microscopy in biology to computational imaging in astronomy, in this talk, I will talk about the problem of nonstationary blind super-resolution, in which the point spread functions associated with point sources need to be calibrated. To do this, I propose a flexible atomic norm minimization framework to solve this daunting inverse problem. Along the way, I also derive a sample complexity bound that is optimal for this problem. This optimization framework also inspires new sensing strategies for modal analysis in structural health monitoring.
At the end of the talk, I will also discuss my summer internship project at Technicolor Research. Motivated by the business at Technicolor, I will explore the possibility of using deep neural networks for standard dynamic range to high dynamic range wide color conversion. Data collection, deep neural network models training and testing, and experimental results will be presented.