小波-Contourlet与迭代Cycle Spinning相结合的SAR图像去噪

De-noising of SAR Images Based on Wavelet-Contourlet Transform with Recursive Cycle Spinning

通过分析合成孔径雷达图像的相干斑噪声模型,提出一种小波-Contourlet与迭代Cycle spinning相结合的SAR图像去噪方法.小波-Contourlet比小波变换、Contourlet变换能更稀疏地表达图像,更好地获得图像结构特征.Contourlet变换缺乏移不变性,导致小波-Contourlet也是缺乏移不变性的,对系数进行阈值处理会产生伪吉布斯现象.Cycle spinning算法可以有效地减少伪吉布斯现象,但不是最优的.为此,用小波变换代替LP(Laplacian pyramid)变换作子带分解,以迭代Cycle spinning代替多次移位取平均值.仿真结果表明,该方法不仅可以显著去除相干斑噪声,达到较高的峰值信噪比,而且还保留了图像的细节,改善了视觉效果.

By analyzing a speckle model of synthetic aperture radar （SAR）, a de-noising method for SAR images based on the wavelet-Contourlet transform and recursive cycle spinning is presented. Compared with wavelet transform and Contourlet transform, wavelet-Contourlet transform can express images more sparsely and better obtain image structure. Because the Contourlet transform lacks shift invariance, wavelet-Contourlet transform also lacks shift invariance. Threshold processing on the coefficients may produce pseudo Gibbs phenomena. Although a cycle spinning algorithm can reduce the pseudo Gibbs phenomena, it is not the best. In this paper, wavelet transform is used to replace the Laplacian pyramid transform （LPT） for sub-band decomposition. Recursive cycle spinning is used to replace the cycle spinning. Simulation results show that the proposed algorithm is efficient, and it performs significantly better in reducing speckle noise, resulting in higher peak signal-to-noise ratio, more image details and better visual quality.