Adaptive multiresolution image and video compression and pre/post -processing of image and video streams
This thesis is divided into two sections. In the first section, the focus is on adaptive transform-based image compression and motion compensation at low bit rates. In the second section, the pre-processing and post-processing of images and video streams are focused on.^ Natural images are two dimensional signals with unknown or time-varying characteristics. For this type of signal, linear expansion with a fixed set of basis functions is not flexible enough to represent the data with the desired degree of accuracy. For example, the Fourier transform is not a good fit for regions with sharp discontinuities such as edges, and the wavelet transform is not a good fit for regions with periodic high-frequency components such as localized textures or stripes. In the first section a new adaptive algorithm for image representation and coding is introduced. This algorithm is based on the concept of segmented orthogonal matching pursuits (SOMP), and adaptively selects the best representation from an overcomplete dictionary of wavelet functions.^ In the second section, a new robust nonlinear filter based on the theory of generalized maximum likelihood and order statistics (GMLOS) is introduced. It is shown that this filter is an $l\sb2$-optimal order statistics filter and some of its properties are proved. A novel algorithm based on wavelet decomposition, variable size kernel GMLOS filters, and soft thresholding for removing the blocking effects in block-based transform coding techniques is introduced. Finally, a simple algorithm for cell packing in ATM networks is introduced, and a novel algorithm for error concealment of images and video streams, based on Multi-directional Recursive Nonlinear Filtering (MRNF) with GMLOS filters is introduced. ^
Major Professor: Rangasami L. Kashyap, Purdue University.
Engineering, Electronics and Electrical
Off-Campus Purdue Users:
To access this dissertation, please log in to our