Denoising of single-look SAR images based on variance stabilization and nonlocal filters

Modeling the distribution of patches with shift-invariance: Application to SAR image restoration

“…We observe that our method does not present the common artefact encontered when using a logarithmically transformed method (pixels darker than their neighbors circled in red) and preserves edges which are blurry without the shift invariance property (areas circled in orange). Figure 4 compares our method with state of the art methods such as [18], [19] and confirms the gain of a shift invariant model.…”

“…Although covariance Figure 3. Comparisons of denoising results between (b) the NL-SAR filter [18] (c) the BM3D filter on log-tranformed data [19], and (d) the proposed approach.…”

NL-SAR: A Unified Nonlocal Framework for Resolution-Preserving (Pol)(In)SAR Denoising

“…• The homomorphic approach, used to extend the NL-means [31], [32] and BM3D [33], first applies a logarithmic transform to the data so that noise becomes additive, then performs a standard non-local filter, and finally applies an exponential transform to map the filtered data back to their original dynamic range. A bias-correction step is necessary to correct for non-Gaussianity in log-space [34].…”

Prediction of Despeckling Efficiency of DCT-Based Filters Applied to SAR Images

“…Due to this, a considerably larger amount of filtering techniques has been designed for removal of AWGN than multiplicative noise. Taking this into account, images corrupted by multiplicative and other types of signal dependent noises are frequently processed applying homomorphic and/or variance stabilizing transformations [32,33]. These transformations for the case of pure multiplicative noise are of logarithmic type [1,25,32] [32], additive noise variance becomes equal to unity although additive noise is also nonGaussian.…”