研究目的
To propose a new method for polarimetric synthetic aperture radar (PolSAR) denoising by addressing a new statistical approach for weights computation in nonlocal approaches, using M-estimators to detect similar pixels and improve covariance matrix estimation.
研究成果
The proposed M-NL method outperforms NL-SAR in most performance measures for PolSAR denoising, showing better smoothing of homogeneous areas, edge preservation, and higher contrast. It offers simplicity in implementation by leveraging M-estimators' properties, making it a robust alternative for speckle reduction in PolSAR imagery.
研究不足
The method assumes specific statistical models (e.g., CES distributions) and may not perform optimally in all non-Gaussian scenarios. Computational complexity could be high due to iterative processes over multiple parameters. Real data evaluation is limited to visual analysis without ground truth.
1：Experimental Design and Method Selection:
The method involves using M-estimators for preestimation of covariance matrices and a binary hypothesis test for pixel selection in nonlocal means denoising. It compares patches based on statistical tests and computes weights using an exponential kernel.
2：Sample Selection and Data Sources:
Simulated PolSAR images generated using Markov random fields and Gibbs distribution, and real RADARSAT-2 PolSAR data of San Francisco Bay.
3：List of Experimental Equipment and Materials:
Computational tools for image processing and analysis; no specific hardware mentioned.
4：Experimental Procedures and Operational Workflow:
Preestimation with Student's M-estimator, pixel selection using Box's M-test, weight computation, and bias reduction. Algorithm iterates over window sizes, patch sizes, and scales.
5：Data Analysis Methods:
Performance evaluated using relative errors on radiometric parameters, correlation parameters, incoherent decomposition parameters (entropy, anisotropy, mean alpha angle), polarimetric signatures, and edge preservation metrics.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
