Abstract:Polarimetric Synthetic Aperture Radar (PolSAR) occupies an important place in remote sensing because it provides richer information about the targets and earth surface compared with single-channel SAR systems.However,PolSAR data is contaminated by speckle noise due to the coherent imaging mechanism,which considerably affects the accuracy of target classification and recognition.Therefore,speckle-noise filtering of PolSAR images is a crucial pretreatment.Nonlocal(NL) means compute the weights between two pixels with similar surrounding neighborhoods (known as patches) instead of two individual pixels.Considering that patches contain structural information,the NL mean filter preserves repetitive structures and performs better than other filters.The key point of the NL algorithm is the similarity criterion setting or the patch weights.This paper proposes a technique to reduce speckle noise using NL means by combining structure and homogeneity similarity.First,image heterogeneity is measured based on the distance of K distribution and is further utilized to distinguish homogeneous and heterogeneous regions.In PolSAR imagery,backscattering from point targets is significantly different from that of distributed media.Strong backscattering from point targets is caused by strong elementary scatterers within a resolution cell.They lack the typical characteristics of speckle and are not random in nature.The preservation of signatures from strong point targets and manmade structures is desired for image interpretation and other applications.In this paper,various samples are collected based on scene heterogeneity.A threshold is utilized to preserve the point and line targets.Then,a new strategy is presented to adapt to the changes in the heterogeneity of the image,which sets the weights of the NL means that were implemented between patches based on the heterogeneity coefficient.Finally,the filtered image is computed.The obtained filter is compared with the refined Lee,mean shift,NLLee,and WisNLTV filters.The qualitative and quantitative aspects of the filters were compared.To compare the ability of the filters to maintain details,corresponding areas in the enlarged span images are shown after filtering with various methods.The proposed method is significantly better on the global and local scales than the existing methods.Moreover,results of H/A/α decomposition show that the proposed method effectively converges the same scattering mechanism and retains complicated scattering mechanisms.The quantitative assessment verifies the equivalent number of looks (a measure of noise reduction),the edge-preserving index,and polarization information preservation on real images.The proposed method has improved filtering performance.The concept of accounting for the heterogeneity coefficient within the NL means algorithm is implemented.The proposed method filters adaptively based on heterogeneity.In addition,comparative results confirm the advantages of the proposed algorithm on both speckle reduction and detail preservation.

Nonlocal Means SAR Image Despeckling Using Principle Neighborhood Dictionaries

A New Nonlocal Iterative Trilateral Filter for SAR Images Despeckling

Sar image despeckling based on nonlocal similarity sparse decomposition

A Novel SAR Image Despeckling Method Based on Local Filter With Nonlocal Preprocessing

Non-Local SAR Image Despeckling Based on Sparse Representation

A SAR Image-Despeckling Method Based on HOSVD Using Tensor Patches.

SAR Image Despeckling by Noisy Reference-Based Deep Learning Method

Nonlocal Means Filtering for Polarimetric SAR Images Based on Heterogeneity

SAR Image Despeckling Based on Block-Matching and Noise-Referenced Deep Learning Method

Patch Ordering-Based SAR Image Despeckling Via Transform-Domain Filtering.

SAR Image Speckle Reduction Based on Nuclear Norm Minus Frobenius Norm Regularization

Sar Image Despeckling By Combination Of Fractional-Order Total Variation And Nonlocal Low Rank Regularization

A Practical Solution for SAR Despeckling With Adversarial Learning Generated Speckled-to-Speckled Images

SAR Image Despeckling Using Quadratic-Linear Approximated L1-Norm

SAR despeckling via classification-based nonlocal and local sparse representation

A Self-supervised SAR Image Despeckling Strategy Based on Parameter-sharing Convolutional Neural Networks

SAR Despeckling Neural Network with Logarithmic Convolutional Product Model

A Boosting SAR Image Despeckling Method Based on Non-Local Weighted Group Low-Rank Representation

A self-attention multi-scale convolutional neural network method for SAR image despeckling

Enhancing RABASAR for Multi-Temporal SAR Image Despeckling through Directional Filtering and Wavelet Transform

SAR Despeckling Using Multiobjective Neural Network Trained with Generic Statistical Samples.