DESPECKLING SAR IMAGES BASED ON STATISTICALLY MODELING LOCAL WAVELET COEFFICIENTS

摘要

A new model for wavelet coefficients of logarithmically transformed SAR images in steerable pyramid is introduced, which models local statistics of underlying coefficients as a Gaussian scale mixture (GSM) model. The model exploits not only intrascale correlation but also interscale dependence. The resulting algorithm significantly reduces speckle in SAR images while preserving the structural features. Simulation results on synthesized images and real SAR images show that our method obtains the best performance both visually and in term of other metrics.