Speckle reduction in polarimetric SAR imagery with stochastic distances and nonlocal means

Abstract

Abstract. This paper presents a technique for reducing speckle in Polarimetric Synthetic Aperture Radar (PolSAR) imagery using Nonlocal Means and a statistical test based on stochastic divergences. The main objective is to select homogeneous pixels in the filtering area through statistical tests between distributions. This proposal uses the complex Wishart model to describe PolSAR data, but the technique can be extended to other models. The weights of the location-variant linear filter are function of the p-values of tests which verify the hypothesis that two samples come from the same distribution and, therefore, can be used to compute a local mean. The test stems from the family of (h-φ) divergences which originated in Information Theory. This novel technique was compared with the Boxcar, Refined Lee and IDAN filters. Image quality assessment methods on simulated and real data are employed to validate the performance of this approach. We show that the proposed filter also enhances the polarimetric entropy and preserves the scattering information of the targets.

Adaptive Local Means Filter for Polarimetric SAR Images; Despeckling for Homogeneous and Heterogeneous Clutter Models

“…Boxcar and Enhanced LEE are at the end of the list regarding the SSIM index. This coincides with the statement of "the most intense blurring produces the best results with respect to ENL" [24]. The non-local mean and the proposed method have similar (with slight improvements in the non-local mean method) values for SSIM and ENL.…”

“…To estimate the noise free value of covariance elements, the maximum likelihood (ML) estimate of C is used, and it is calculated from covariance elements of the all pixels defining the surface [21][22][23][24].…”

“…The data clutter is described by two parameters: the normalized texture and the speckle normalized covariance matrix. As described in the previous sections, for homogeneous areas (case-1), the texture random variable has a constant value in the neighbourhood of a resolution cell, the maximum likelihood (ML) can be used to estimate the free-noise covariance matrix, and is calculated only from all the pixels defining the homogeneous surface [24]. The heterogeneous areas (case-2) can be modelled by some probability density function (PDF).…”

“…To check the validity of the proposed system with respect to texture information as a random variable for non-homogeneous blocks, we compare the proposed method with the algorithm proposed by Leonardo Torres [24]. Torres' method divides the image into patches with fixed size.…”

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

“…It has recently been extended following the principles of the successful BM3D approach with SAR-BM3D [41]. Different statistical tests have been proposed for the pixel selection, including: joint-likelihood criteria [37], [38], [42], generalized likelihood ratio tests [40], [43], stochastic and geodesic distances [26], [44]. Some of them are free of selection bias (see Section 3.8 in [45] for more details).…”