Publication: Research › Ph.D. thesis – Annual report year: 2005
Methods for segmentation and restoration of SAR data using Markov Random Fields (MRF) have been studied extensively by many researchers over the last two decades. What is of special interest is not only methods for segmentation and classification of SAR data for land cover labeling applications, but also methods for detail preservation, which have experienced a rapid growth over the past few years. The main part of this thesis concerns the development of image restoration methods that facilitate the extraction of biotope relevant information from polarimetric SAR data. Because the semi-natural environments under study are very small, it is crucial for this investigation that the restoration methods are capable of restoring fine structures as well as preserving homogeneous areas. The restorations are carried out in a signal adaptive mode using MRF in a Bayesian framework. Different a priori models are implemented in both the local optimizer Iterated Conditional Modes (ICM) and the global optimization technique Simulated Annealing (SA). A new technique for algorithm optimization is presented, which relies on ratios of SAR data and their histograms. A quantitative evaluation of the restorations based on statistics derived from the ratio images is presented together with comparative analyses of restorations using ICM and SA. The relation between the restored polarimetric SAR data and in situ data collected at two semi-natural wetland and grassland areas is investigated using multivariate techniques. The restored polarimetric SAR data are classified by using a supervised and an unsupervised classifier and comparative analyses of their performances are carried out.
|State||Published - Jul 2005|
No data available