【摘要】 将非下采样Directionlet变换(Nonsubsampled Directionlet,ND)和高斯混合尺度(GSM)模型相结合,提出了基于ND-GSM模型的采样矩阵方向优化算法并将其应用于SAR图像去噪。首先,将SAR图像的分割子图进行二进小波变换,从而确定SAR图像的方向优化采样矩阵,然后在各个子图中将GSM模型引入采样矩阵方向优化的非下采样Directionlet变换域中,构造了采样矩阵方向优化的非下采样Directionlet域分解系数的邻域模型(ND-GSM),最后利用Bayes最小均方估计进行子图变换域的局部去噪,并合成去噪后的分割子图,得到去噪后的SAR图像。该方法解决了当非下采样Directionlet基函数的方向与图像中各向异性目标不一致时图像的逼近效果差的问题。仿真实验结果表明,该方法能充分体现邻域间系数的相关性,同时在图像边缘等细节特征保持方面具有明显优势,明显改善了图像视觉效果,取得了比空域滤波及小波方法更优的去噪性能。更多还原

【Abstract】 A new improved algorithm of the sampling matrix direction based on the model of ND-GSM combining with the nonsubsampled Direction lettransform and the model of Gaussian scale mixtures was presented.First,binary wavelet transform is applied to segment subgraphs of SAR image to determine thesampling matrix of optimizing direction of SAR image.Then,by combining nonsubsampled Directionlet transform which optimizes the direction of the sampling matrix with Gaussian scale mixtures in the segmentation subgraphs,the marginal distributions of neighbor coefficients in the nonsubsampled Directionlet domain with the sampling matrix of direction optimization are modeled.Finally,for removing the speckle noise,the Bayes least square estimation is adopted to evaluate each coefficient.The denoised segmentation subgraphs is composed to obtain the SAR image after denoising.The method solves the phenomenon that image approximation effect is bad when the direction of nonsubsampled Directionlet basis function and the direction of the image anisotropic target is inconsistent.Simulation results show that theme thod can fully reflect strong correlation among the amplitudes of neighbor coefficients,have obvious advantage in image detail preservation,improve the image visual effect and achieve a better denoising performance than the spatial filtering and wavelet method.更多还原