Calendar

ECE Seminar, Singanallur V. Venkatakrishnan Singanallur V. Venkatakrishnan Lawrence Berkeley National Lab Berkeley, California Faculty Host: Siddharth Ramachandran Light refreshments will be available outside of PHO 339 at 10:45 am Title: Computational Sub-micron Imaging: a Model-based Iterative Reconstruction approach Abstract: High-resolution imaging instruments such as electron microscopes and X-ray microscopes play a critical role in addressing grand-challenges in fields ranging from advanced manufacturing to developing treatment strategies for various diseases by enabling the characterization of samples at the sub-micron scale. Driven by new imaging experiments that aim to characterize materials and phenomena at a high spatial and temporal resolution in 2D or 3D, these instruments are producing large amounts of data that has to be computationally inverted in order to reconstruct images and extract the relevant information. While advances have been made in improving the hardware via better optics and detectors, the algorithms typically used for computationally forming the images have not fully exploited the underlying models associated with the measurement, thereby resulting in sub-optimal reconstruction quality. In this talk, Venkatakrishnan will address the “algorithmic gap” in computational sub-micron imaging by using a model-based iterative reconstruction (MBIR) approach. MBIR, based on a Bayesian estimation framework, provides a systematic framework for combining the physics of image formation and the statistics of the measurement (the forward model) with models for the underlying object (the prior model) to address the inversion as minimizing a unified cost-function. First, Venkatakrishnan will present an MBIR algorithm for nano-tomography using an electron microscope. Venkatakrishnan will demonstrate how the challenges of a non-traditional acquisition geometry, missing calibration parameters and presence of measurement anomalies due to the nature of certain samples can be overcome to dramatically improve the quality of 3D reconstructions compared to the current art, thereby expanding the capability of this powerful instrument. In particular, Venkatakrishnan will discuss the formulation of a new forward model and the subsequent the design of an optimization algorithm based on a majorization-minimization approach to find a solution to the reconstruction problem. Next, Venkatakrishnan will show how the MBIR approach can also significantly improve the imaging capability of synchrotron based X-ray microscopes. Finally, Venkatakrishnan will address an important aspect of enhancing MBIR - the ability to use sophisticated image (prior) models to improve reconstruction quality for a given application. Venkatakrishnan will present a framework, called “Plug-and-Play” (P&P) priors, which can enable new image models, defined via de-noising algorithms, to be conveniently used for model-based inversion. Venkatakrishnan will present results of adapting the P&P framework for sub-micron imaging applications. Bio: S. V. Venkatakrishnan received a B.Tech. degree in electronics and communication engineering from the National Institute of Technology Tiruchirappalli in 2007, and an M.S. degree from the School of Electrical and Computer Engineering at Purdue University in 2009. From 2009 - 2010 he was employed as a Research and Development engineer at Baker Hughes Inc. where Venkatakrishnan developed signal-processing algorithms for logging-while-drilling imaging tools. In 2014, he received a Ph.D. from the School of Electrical and Computer Engineering at Purdue University. Venkatakrishnan was awarded a Presidential Scholar Award at the Microscopy and Microanalysis conference (2014) for his work on the development of an algorithm for low-dose electron tomography. Venkatakrishnan is currently a post-doctoral fellow at Lawrence Berkeley National Lab affiliated with the Advanced Light Source and the Center for Applied Mathematics for Energy Research Applications. His research interests include Computational Imaging, Inverse Problems, Statistical Signal Processing and Machine Learning.