Despeckling techniques for polarimetric sar images using bandelet transform

Abstract

Satellite-based earth observation data became the primary source of a wide variety of newlineapplications in hydrology, agriculture, oceanography, forestry, disaster monitoring, and newlinemilitary surveillance. The last three decades have witnessed stupendous growth in the newlinesatellite-based applications because of the advanced Synthetic Aperture Radar based newlineimaging. Even though the end users are more conversed with optical images, a quantum newlinejump of acceptance and usage of satellite-based SAR images was seen due to its all- newlineweather, day and night imaging capability. Polarimetric SAR added an extra dimension newlineto SAR advantages with polarization diversity in imaging enabling to derive the target newlinefeatures accurately. In contrast to conventional SAR, polarimetric SAR detects the newlinetargets by using multiple polarizations. Similar to SAR images, PolSAR images are newlineprone to multiplicative speckle noise which degrades image quality due to its coherent newlineimaging mechanism. Hence despeckling is an essential pre-processing step for newlineeffectively utilizing the PolSAR images for remote sensing applications. newlineThe first part of the research concentrated on developing a transform domain newlinefiltering approach for polarimetric SAR images using bandelet transform as spatial newlinedomain filtering impairs the retrieved image by over smoothing and removal of image newlinefiner details. Multiplicative speckle noise model based on Wishart distribution in newlineconjunction with thresholding of bandelet coefficients is applied to the covariance newlinematrix elements. The GCV thresholding scheme is adopted as the optimum threshold newlineselection scheme. The proposed method is validated through the simulated images newlinecreated using the Montecarlo simulation technique and real polarimetric images newlineacquired by various airborne and spaceborne PolSAR systems.