Nonparametric Edge Detection in Speckled Imagery

This thesis proposes a non-parametri te hnique for boundary dete tion in spe kled imagery. Syntheti Aperture Radar (SAR), sonar, B-ultrasound and laser imagery is orrupted by a signal-dependent non-additive noise alled spe kle. Several statisti al models have been proposed to des ribe su h a noise, thus of spe ialized te hniques for image improvement and analysis. The G distribution is a statisti al model that su eeds in des ribing a wide range of areas as, for instan e in SAR data, pastures (smooth), forests (rough) and urban (extremely rough) areas. The aim of this thesis is to develop alternative te hniques for edge dete tion in spe kled imagery. Its starting point are the works by Gambini et al. (2006, 2008). We des ribe a new edge dete tor based on the Kruskal Wallis test and show that it is an useful alternative to the method proposed by M. Gambini, whi h is based on the likelihood fun tion of the data. We provide eviden e that the M. Gambini te hnique an be su essfully repla ed by the Kruskal Wallis method. The latter is more omputationally eÆ ient, the orresponding algorithm being up to 1000 times fasted that the M. Gambini algorithm.