An experimental setup for multiresolution despeckling of COSMO-SkyMed image products

This paper describes the most recent achievements in speckle reduction of COSMO-SkyMed (CSK ®) synthetic aperture radar (SAR) data. An advanced multresolution despeckling filter, based on undecimated wavelet transform (UDWT) and maximum a-posteriori (MAP) estimation has been specialized and optimized to CSK ® data, both single- and multilook. The tradeoff between performances and computational complexity has been investigated: Laplacian-Gaussian and generalized Gaussian (GG) priors for MAP estimation in UDWT domain differ by one order of magnitude in computation cost. The former are more complex but yield the best results attainable with a Bayesian estimation carried out in the UDWT domain. Pre-processing of point targets and segmentation of wavelet planes has been exploited to effectively handle the heterogeneity of the data. The effects of multilooking have been investigated. Starting from single-look complex (SLC) data, the spatial correlation coefficients (CC) of speckle and the equivalent number of looks (ENL) of all products have been theoretically calculated. It is proven that, besides having an inherently better radiometric quality, multilooked products exhibit a lower spatial correlation of speckle than single-look products, thereby better falling under the assumption of uncorrelated speckle, exploited by the majority of model-based despeckling filters, included those used in the present work. The effects of spatial resampling have been investigated as well. Unlike MAP filters in spatial domain (e.g. the Gamma-MAP filter), MAP filters in wavelet domain are little sensitive to resampling, because the fundamental hypotheses on which they rely are not violated because of resampling. Comparisons with the state of the art are also provided and shown to be more than favorable. Besides traditional supervised methods to evaluate the quality of despeckling, a novel procedure, fully automated, based on bivariate analysis of noisy and denoised image has been devised. Its results agree both with visual analysis and with manual measurements.