Files in this item



application/pdfNguyen_Tan.pdf (2MB)
(no description provided)PDF


Title:Computation tools for the Fourier transform infrared (FT-IR) spectroscopic imaging
Author(s):Nguyen, Tan
Advisor(s):Do, Minh N.; Bhargava, Rohit
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Fourier transform infrared spectroscopy (FT-IR)
sparse reconstruction
compressive sensing
hyperspectral data
Abstract:Over the pass 60 years, infrared (IR) spectral imaging has become an important tool with various applications such as identifying chemical distributions of biomedically relevant components in tissue, investigating biomedical and biological processes, and studying development of diseases (histopathology). This thesis focuses on denoising and deblurring absorbance images of the Fourier transform infrared spectroscopy for improved spatial resolution while maintaining spectral quality. In addition, it aims to speed up data-acquisition by deploying state-of-the-art computational tools such as dictionary training for sparse representation, and compressive sensing. Here, we use a singular value decomposition denoising algorithm to recover the noiseless absorbance data. Then, novel variational Bayesian deconvolution algorithms using a theoretical formula of the optical point spread function (PSF) are used to improve the spatial resolution, and estimate the mismatching term in the true and theoretical PSF. For sparse reconstruction, we train a K-SVD dictionary to sparsely represent the interferograms. Then, using optimization algorithms, we recover the full dimensional interferograms from very few measurements. Using experimental results on the standard United States Air Force (USAF) 1951 target and breast tissue samples, we show an improvement of 10.53 dB in the signal-to-noise ratio (SNR) after denoising. In addition, the absorbance contrast ratio (ACR) is increased by at least 1.07 times after deblurring over a spatial frequency range of interest on the standard USAF target. Most importantly, our method improves the spatial resolution without significantly modifying the underlying spectral information. For sparse reconstruction, we demonstrate that reconstruction results with correlation factors of at least 0.999, and mean relative errors as small as 3% can be obtained by using just 32 measurements (1.9% of the total number of measurements).
Issue Date:2012-05-22
Rights Information:Copyright 2012 Tan Nguyen
Date Available in IDEALS:2012-05-22
Date Deposited:2012-05

This item appears in the following Collection(s)

Item Statistics