University Events

Description: Topic: A Comparison of Thresholding Methods for Two-Dimensional Wavelet-Based Image Denoising

Location: Charlton College of Business Conference Room, CCB-115

Abstract:
Applications in remote sensing, medical imaging, and target detection rely on relatively noise-free images to draw essential inferences for machine learning and eventual human interaction. Denoising images can be challenging due not only to the stochastic nature of additive noise but also to the complex, varied, and unpredictable nature of the actual image to be recovered. For these reasons, image-processing techniques are developed to provide solutions to improve images taken from corrupted measurements.

The Discrete Wavelet Transform (DWT) is widely used within the image processing community as it provides a bases that concentrates energy in relatively few coefficients for a wide range of image types. Such representations are considered sparse and do not lend themselves to conventional linear filtering. The DWT provides a computationally fast projection of the image onto the orthogonal multiscale and spatially localized wavelet bases. Unlike the conventional Fourier bases, which rely on phase across the entire set of bases to capture localized features, the DWT captures the scale and localized nature of the image features in a very small subset of bases. This particular property of wavelets proves useful as many two-dimensional signals prove quite sparse in the wavelet domain. Optimal denoising of such sparse, non-stationary structures requires the adaptation of conventional methods.

This thesis focuses on denoising images via nonlinear thresholding techniques in the wavelet domain. Different adaptive and non-adaptive wavelet-based thresholding methods are presented: VISU-Shrink, SURE-Shrink, and Bayes-Shrink. For clarity, the focus is provided on four test images corrupted by additive white Gaussian noise (AWGN) at several noise variance levels. The relative merits of soft and hard thresholding are also explored. Performance in terms of mean-squared error, peak-signal-to-noise ratio (PSNR), and structural similarity index are provided. Under these test cases, Bayes-Shrink soft thresholding outperformed VISU-Shrink and SURE-Shrink. Bayes-Shrink soft thresholding, on average, generates an improvement of PSNR of roughly 14.5 dB.

NOTE: All ECE Graduate Students are ENCOURAGED to attend.
All interested parties are invited to attend. Open to the public.

Advisor: Dr. Paul J. Gendron
Committee Members:
Dr. John R. Buck, Chancellor Professor, Department of Electrical & Computer Engineering, UMASS Dartmouth; Dr. Antonio H. Costa, Professor, Department of Electrical & Computer Engineering, UMASS Dartmouth

*For further information, please contact Dr. Paul J. Gendron via email at pgendron@umassd.edu.