Speckle Reduction in Synthetic Aperture Radar Images in Wavelet Domain Using Laplace Distribution

Speckle is a granular noise-like phenomenon which appears in Synthetic Aperture Radar (SAR) images due to coherent properties of SAR systems. The presence of speckle complicates both human and automatic analysis of SAR images. As a result, speckle reduction is an important preprocessing step for many SAR remote sensing applications. Speckle reduction can be made through multi-looking during the image formation or using spatial filters as a preprocessing step. However, these methods have some limitations such as a decrease in spatial resolution or smoothening of details and edges. To overcome these problems, Multi-Resolution Analysis (MRA), such as wavelet transform, should be used. In this paper, a despeckling method based on the Bayesian theory and Maximum a Posteriori (MAP) estimator in the wavelet domain was proposed. The noise-free wavelet coefficients of the logarithmically transformed image and the noise in the wavelet domain were modeled based on the Laplace and Gaussian distributions respectively. VisuShrink, SureShrink, and BayesShrink methods were also implemented and applied to both simulated and real SAR data for comparison purpose and to assess the proposed method. PSNR and beta edge preserving index were used to evaluate the performance of simulated SAR data, while ENL was employed to evaluate the real SAR data. Experimental results of despeckling showed the superior performance of the proposed method in suppressing the speckle efficiently and preserving better the spatial details in the SAR image.