基于组稀疏混合模型的遥感图像去噪方法

Remote Sensing Image Denoising Method Based on Group Sparse Mixture Model

摘要: 在遥感图像的拍摄和传输过程中，会产生大量的噪声，高斯噪声和椒盐噪声是较为常见的两种噪声，目前的去噪算法对于这类混合噪声的去除普遍存在边缘模糊等问题。针对此问题，文章提出了一种新的基于组稀疏混合模型的遥感图像混合噪声的去除方法，首先通过双边矩阵乘法提高块组的稀疏性，然后通过块组独立这一假设提出了基于块组的混合噪声去噪框架，接着对辅助变量、估计的图像、椒盐噪声分别进行最小化问题的优化求解，最后通过聚合块组得到去噪后的图像。实验结果表明，本文的算法能够有效地去除遥感图像中的高斯噪声和椒盐噪声，相对于其他传统方法具有更高的PSNR、SSIM以及FSIM数值。

Abstract: In the process of capturing and transmitting remote-sensing images, a large amount of noise is generated. Gaussian noise and salt-and-pepper noise are two common types of noise. Current denoising algorithms generally have problems with edge blurring when removing such mixed noise. To address this problem, we propose a new method based on a group sparse mixed model to remove mixed noise in remote sensing images. Firstly, the sparsity of patch groups is improved through bilateral matrix multiplication. Then, a patch group-based mixed noise denoising framework is proposed based on the assumption of patch group independence. Then, the auxiliary variables, estimated images, and salt and pepper noise are optimized and solved separately. Finally, the denoised image is obtained by aggregating patch groups. Experimental results show that the algorithm in this paper can effectively remove Gaussian noise and salt-and-pepper noise in remote sensing images and has higher values of PSNR, SSIM, and FSIM compared with several popular algorithms.