Due to the limit of imaging system, synthetic aperture radar images are often corrupted with non-Gaussian multiplicative speckle noise. For the purpose of noise suppression, an algorithm called NLM-TV is proposed, which integrates non-local means filter and total variation regularization. It involves three steps. First, convert the multiplicative noise into signal-dependent additive noise, dividing the image into three categories based on noise level, the edge, the strong scattering region and the weak scattering region. Non-local means filter was then applied. To maintain the edge structure effectively, the smooth parameter must be small. In the strong scattering region, TV regularization was used because of insufficient reduction of speckles. In this paper, several simulations were conducted in RADARSAT-2 and TerraSAR-X images, the results showed that compared to a variety of filtering algorithms, NLM-TV algorithm could significantly increase the equivalent number of looks both in the weak scattering area and strong scattering regions. At the same time, edge keeping index could be increased by more than 10%.%受成像体制影响，合成孔径雷达（Synthetic aperture radar，SAR）图像带有非高斯的乘性相干斑噪声。为有效抑制乘性相干斑噪声，提出一种融合非局部均值滤波（Non-local means filter, NLMF）与全变差（Total variation，TV）正则化的非局部均值-全变差（NLM-TV）降噪算法。首先将相干斑噪声转换为依赖于散射强度的加性噪声，将SAR图像分为边缘、强散射区、弱散射区。然后利用NLMF进行降斑，为有效的保持边缘结构， NLMF的平滑参数选取较小。在强散射区，为解决平滑参数较小所带来的降斑不充分问题，进一步使用TV正则化进行平滑处理，获得最终的降噪结果。使用RADARSAT-2，TerraSAR-X两景实测SAR图像仿真实验，结果表明：相比多种滤波算法， NLM-TV算法在弱散射区，强散射区均能显著提高等效视数，边缘保持指数能够提高10%以上。
展开▼
