SAR Image Denoising Via The Estimation Of Sparsity Coefficients In Transform Domain

Due to the unique advantages of all-time, all-weather, strong penetration and high-resolution imaging, Synthetic aperture radar(SAR) plays an irreplaceable role in various fields of national economy and national defense. However, SAR images are inherently affected by speckle noise, which is due to the principle of coherent imagery. The presence of speckle noise has a crippling effect on the following recognition and interpretation of SAR images. Thus the speckle noise reduction has become a key step of the SAR image processing. In recent years, the transform-domain sparse representation-based denoising methods have attracted widespread attention. With the assume that the noise-free signal is sparse in the transform-domain, image denoising can be effectively implemented by preserving the few high-magnitude transform coefficients which represent the true-signal and discarding the rest which convey mostly noise energy. Taking the estimation of transform-domain sparse coefficients as main line and combining statistical characteristics of SAR images, this paper mainly did the research on the reduction of speckle noise in SAR images and the main work of this paper is as follows:(1) Based on some research on the principle of transform-domain sparse representation-based denoising and the statistical characteristics of SAR images, some core problems of transform-domain sparse representation-based denoising are proposed: how to choose the proper dictionary to represent image more sparsely, how to estimate the sparsity coefficients more accurately, how to construct the sparse representation-based SAR image denoising model and how to solve the proposed model effectively and efficiently.(2) First, an effective method is proposed for transforming image into transform-domain by using shearlet. Second, the stagewise orthogonal matching pursuit(St OMP) is used to obtain the optimization solution in transform domain. Finally, the projected adaptive total variation scheme is used to make up the loss of the image details result from the coefficients dropped in the sparse representation processing.(3) A novel transform-domain sparse representation-based denoising method is presented based on the combination of sparse representation and nonlocal self-similarity, and the linear minimum mean square error(LMMSE) was used to estimate the noise-free transform domain coefficients more accurately based on the statistical characteristics of SAR images.(4) Different regions of SAR image not only have different statistical characteristics but also have different denoising goals, hence a classification-based denoising method is proposed to suppress speckle noise, which exploits 3D transform-domain threshold shrinkage and nonlocal weighted average to deal with heterogeneous regions and homogeneous regions, respectively.