Some Estimation and Restoration Techniques for Statistical Image Analysis

Statistics

Archer, Graeme Ernest Barclay

University:University of Glasgow

关键词: Statistics;   

Others  :  http://theses.gla.ac.uk/76438/1/13834220.pdf

来源: University of Glasgow

【 摘 要 】

This thesis is concerned with statistical image analysis: the estimation of parameters within image models, and how to produce restorations of degraded scenes which are the most probable given the parameter estimates and the data. We develop algorithms for estimation within hierarchical and empirical Bayesian models, and compare results with non-Bayesian methods. The empirical behaviour of parameter estimates under different algorithms are studied in a simulation exercise and compared with their theoretical behaviour. We sample realisations from Markov random fields using the Metropolis algorithm, and propose a resampling technique to assess convergence. An alternative to the EM algorithm (EMA), the Image Space Reconstruction algorithm (ISRA), is extended and compared with the EMA. A technique for increasing the rate of ISRA-convergence is investigated. Finally, an adaption of a method to prevent over-smoothing of image discontinuities is fully automated. The effect of user-supplied parameter values on the image restoration quality is investigated via a simulation study; the effects are found to be negligible.