Detection and estimation in speckled images based on marked point process speckle noise models
Abstract
Speckle noise arises in coherent imaging systems such as synthetic aperture radars, biomedical ultrasound systems, laser imaging systems, and imaging sonar systems. Detection, pattern recognition, and estimation in images corrupted by speckle is complicated by the nature of the noise. In this work, we examine these problems, with an emphasis on speckle as it arises in synthetic aperture radars. Our work on detection focuses on developing optimal detectors for general pattern recognition or discrimination problems in speckled images, characterizing their performance, and making comparison with models of human detection of objects and features in fully developed speckle. This work was motivated by the increasing need for the development of machine related detection algorithms for automated SAR image analysis, especially with the recent increase in the volume of SAR data collected and the resulting complexity of data analysis by humans. In our work on estimation, we consider partially developed speckle, which can be viewed as a carrier of useful surface roughness characteristics as opposed to that of a contaminating noise. Our goal is to estimate the scattering density and average surface reflectivity of SAR images from SAR intensity measurements. These estimation techniques provide one possible approach to characterizing scattering surfaces. We first studied the statistics of partially developed speckle based on a doubly stochastic marked Poisson point process. Using this statistical model, we derived an expectation-maximization algorithm to produce a recursive optimal maximum likelihood estimate of the average surface reflectivity of SAR images. This was done for the particular model of a homogeneous surface with a constant mean reflectivity within a resolution cell. We then studied the effect of SAR system parameters on the estimator performance.
Degree
Ph.D.
Advisors
Bell, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.