Multiresolution decompositions and models for signal processing applications.

Format:

Book

Thesis/Dissertation

Author/Creator:

Anandakumar, Krishnasamy.

Contributor:

University of Pennsylvania.

Language:

English

Subjects (All):

Electrical engineering.

0544.

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

Local Subjects:

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

0544.

Physical Description:

204 pages

Contained In:

Dissertation Abstracts International 58-11B.

System Details:

Mode of access: World Wide Web.

text file

Summary:

Multiresolution signal decomposition provides a very effective framework for signal representation. In multiresolution decomposition, a signal is interpreted as a set of detail signals which appear at different resolutions. In linear multiresolution decomposition schemes, the resolution of a signal is linked to its frequency content and hence each detail signal corresponds to a particular frequency band of the input signal. In general, the detail signals are treated independently, as different detail signals correspond to different frequency bands. Such frequency-based treatment ignores the spatial characteristics of the signal components. In this dissertation we consider multiresolution schemes that exploit both spatial and frequency properties of the signal components.

In the first part of the dissertation we present wavelet-based self-affine signal models that exploit similar components present in typical signals. The analysis of the proposed self-affine models leads to a relationship among the wavelet coefficients of the modeled signal. This relationship shows that the signal details at finer resolutions can be predicted from the signal details at coarser resolutions. The effects of our self-affine modeling procedure in the presence of modeling error are thoroughly analyzed and discussed. After considering the effects of model parameters on reconstructed signals, we propose a wavelet-based self-affine scheme that has a noniterative optimal modeling procedure. We study the applications of the proposed wavelet-based self-affine models to gray-scale image compression and speech compression. We also present multivariate wavelet-based self-affine models that exploit both inter and intra component redundancies present in the input multivariate signals.

The second part of this dissertation deals with order statistic based nonlinear multiresolution schemes that decompose the signals according to their spatial and frequency characteristics. These schemes generalize the concept of resolution by decomposing stationary components from the frequency content, while extracting edges and sharp features according to their sizes (for one-dimensional signals) or structural properties (for two-dimensional signals). The deterministic and statistical properties of the multiresolution decomposed signals are considered in detail. As an application of the proposed schemes, we study the restoration of images in the presence of Gaussian and non-Gaussian noise.

Notes:

Thesis (Ph.D. in Electrical Engineering) -- University of Pennsylvania, 1997.

Source: Dissertation Abstracts International, Volume: 58-11, Section: B, page: 6118.

Local Notes:

School code: 0175.

ISBN:

9780591659467

Access Restriction:

Restricted for use by site license.