Back to search

National Technical University of Athens (NTUA)

Μέθοδοι επεξεργασίας πολυμεσικών σημάτων για βέλτιστη αποθορυβοποίηση και ανίχνευση με εφαρμογές σε δορυφορικά δεδομένα

Abstract

dc:description

Our work lies in the area of statistical signal and image processing. Our main interest is the development of optimal signal processing methods for denoising and detection. In the current work we follow two research directions. The first one is focused on the problem of denoising images which suffer from signal-dependent noise and is motivated by the problem of enhancing the quality of satellite images degraded by quantum noise. The second direction is related to the study and development of multi-sensor signal denoising and detection methods. In the first part of the thesis we study the problem of photon-limited image denoising. In many modem imaging systems image acquisition is accomplished by detecting photon counts at different spatial locations of a sensor. For low intensity levels, one of the dominant noise sources responsible for the degradation of the quality of the captured images is the so-called quantum or shot noise. This kind of noise is due to fluctuations on the number of detected photons, an inherent limitation of the stochastic nature of the detection process. Quantum noise obeys a Poisson law and, hence, is highly dependent on the underlying intensity pattern being imaged. The approach we follow in the thesis to deal with the problem of estimating the underlying image intensity is based on a representation of the image over multiple scales, using a family of multiscale nonlinear transforms. The representation of the image in the new domain permits us to employ graphical models in order to capture its statistical properties and can lead to efficient intensity estimation schemes. In this framework, we discuss important statistical properties of natural images in the domain of the transform, captured to a good extent by the family of multiscale Markov tree models which we study in depth. One of our main contributions in this area is the development of a novel multiscale nonlinear transform which can decouple the Poisson noise and the image true content. In addition, the proposed transform is proved that is closely related to the wavelet transform. This relation is of significant importance since it implies that key wavelet properties are also inherited by the proposed transform. Moreover, we propose new algorithms for learning the statistical model parameters directly from noisy data. The accurate estimation of the model parameters is crucial since it can lead to improved denoising results. In the second part of the thesis we study multi-sensor signal processing methods. Signals that are captured separately by a sensor of the array can be considered as different realizations of the same source signal. The goal in multi-sensor signal processing is to combine these realizations and exploit their correlation in a way that will lead to an output signal of superior quality compared to the input signals. In this research area we study statistical optimal multi-sensor denoising methods to deal with the problem of signal enhancement. The studied methods correspond to multichannel filtering of the captured signals. However, despite their theoretically optimal results, these filters are difficult to realize. This is due to the requirement for knowledge of second order statistics for both the signal and the corrupting noise that makes these filters signal-dependent. In order to address this issue, we propose a novel generalized estimation method for estimating the necessary parameters. This method is general, appropriate for different noise conditions and various optimization criteria. Finally, we also study the problem of automatic signal detection and recognition under noisy conditions. Our main research contributions in this area is the development of a theoretical unified framework of two basic energy operators, which are used for obtaining energy measurements and the development of a novel multi-sensor feature extraction scheme which can serve as the front-end of a statistical recognition system.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2009

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Lefkimmiatis, Stamatios
  • Λευκιμμιάτης, Σταμάτιος

Subjects

dc:subject × 21

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/18846

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Lefkimmiatis, Stamatios; Λευκιμμιάτης, Σταμάτιος. Μέθοδοι επεξεργασίας πολυμεσικών σημάτων για βέλτιστη αποθορυβοποίηση και ανίχνευση με εφαρμογές σε δορυφορικά δεδομένα. National Technical University of Athens (NTUA), 2009. http://hdl.handle.net/10442/hedi/18846