Image Despeckling Based on Improved LMMSE Wavelet Shrinkage

Abstract

The automatic interpretation of SAR images is often extremely difficult due to speckle, a signal dependent noise, which is inherent of all active coherent imaging systems. Thus, despeckling has become a crucially important issue in SAR image processing. Wavelet theory provides a powerful tool for detecting image feature at different scales. Wavelet-based algorithms have been widely used to reduce speckle noise. In this paper, an adaptive despeckling method for synthetic aperture radar (SAR) images is proposed based on wavelet shrinkage. It follows the framework of the linear minimum mean square error (LMMSE) filter in the wavelet domain proposed for speckle suppression, but improves the parameter estimation method by taking into account the distribution property of wavelet coefficients based on the bilateral kernel regression. An improved adaptive shrinkage function is obtained and each coefficient is decided separately. Simulation results for the simulated SAR images demonstrate the proposed modified method outperforms some representative SAR despeckling methods when the noise is not serious. Keywords-synthetic aperture radar (SAR); speckle; wavelet transform; linear minimum mean square error (LMMSE); kernel regression