Reconstruction, Classification, and Segmentation for Computational Microscopy

Bayesian Image Reconstruction;Blind Deconvolution;Computational Microscopy: Magnetic Resonance Force Microscopy (MRFM) and Electron Backscatter Diffraction (EBSD);Dictionary-based Classification;Segmentation;and Anomaly Detection;Markov Chain Monte Carlo and Variational Bayes Methods;Uncertainty Quantification and Super-resolution;Electrical Engineering;Engineering;Electrical Engineering: Systems

Park, Se UnNadakuditi, Rajesh Rao ;

University of Michigan

关键词: Bayesian Image Reconstruction;    Blind Deconvolution;    Computational Microscopy: Magnetic Resonance Force Microscopy (MRFM) and Electron Backscatter Diffraction (EBSD);    Dictionary-based Classification;    Segmentation;    and Anomaly Detection;    Markov Chain Monte Carlo and Variational Bayes Methods;    Uncertainty Quantification and Super-resolution;    Electrical Engineering;    Engineering;    Electrical Engineering: Systems;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/102296/seunpark_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This thesis treats two fundamental problems in computational microscopy: image reconstruction for magnetic resonance force microscopy (MRFM) and image classification for electron backscatter diffraction (EBSD). In MRFM, as in many inverse problems, the true point spread function (PSF) that blurs the image may be only partially known. The image quality may suffer from this possible mismatch when standard image reconstruction techniques are applied. To deal with the mismatch, we develop novel Bayesian sparse reconstruction methods that account for possible errors in the PSF of the microscope and for the inherent sparsity of MRFM images. Two methods are proposed: a stochastic method and a variational method. They both jointly estimate the unknown PSF and unknown image. Our proposed framework for reconstruction has the flexibility to incorporate sparsity inducing priors, thus addressing ill-posedness of this non-convex problem, Markov-Random field priors, and can be extended to other image models. To obtain scalable and tractable solutions, a dimensionality reduction technique is applied to the highly nonlinear PSF space. The experiments clearly demonstrate that the proposed methods have superior performance compared to previous methods. In EBSD we develop novel and robust dictionary-based methods for segmentation and classification of grain and sub-grain structures in polycrystalline materials. Our work is the first in EBSD analysis to use a physics-based forward model, called the dictionary, to use full diffraction patterns, and that efficiently classifies patterns into grains, boundaries, and anomalies. In particular, unlike previous methods, our method incorporates anomaly detection directly into the segmentation process. The proposed approach also permits super-resolution of grain mantle and grain boundary locations. Finally, the proposed dictionary-based segmentation method performs uncertainty quantification, i.e. p-values, for the classified grain interiors and grain boundaries. We demonstrate that the dictionary-based approach is robust to instrument drift and material differences that produce small amounts of dictionary mismatch.