针对小波阈值法在去除遥感图像高斯噪声时,所存在的由于过度"扼杀"小波系数而引起的模糊边缘问题,以及P-M模型通常会使图像的灰度趋于分段常量而产生所谓的"块状"效应问题。提出小波域偏微分方程(PDE)遥感图像去噪模型,该模型通过对遥感图像进行小波分解,保持低频子带信息,而只对含有噪声、图像边缘的高频子带进行基于子带方向特性的非线性异性扩散,使模型在有效去除高斯噪声的同时,能够很好地保护遥感图像中的边缘特征和细节纹理信息,避免了去噪后的结果图像出现分段常量现象。实验结果表明,对于相同的遥感图像高斯噪声,基于所提出混合模型的去噪图像的PSNR较基于类零树的Bayes阈值法和P-M模型提高了1~2dB。

The noise analysis and elimination in remote sensing images has attracted considerable attention,and has become an important research field for remote sensing image processing. In this paper,we propose a wavelet threshold method to de-noise the Gaussian noise in remote sensing image to make the edge fuzzy causing by over the existence of the "strangulation" of the wavelet coefficients,as well as P-M model usually tends to make the image gray sub-constant,resulting the so-called "massive" effect problem,This paper proposes a new remote sensing image denoising model based on wavelet partial differential equations (PDE) to address the above mentioned issue.This model decomposes remote sensing images by wavelets and maintain the low-frequency subband information.Only with noise and the edge's high-frequency sub-band based on sub-band directional characteristics of the nonlinear anisotropic diffusion,this model can remove Gaussian noise well and,at the same time,can also protect the edge features and details of remote sensing image,and avoids to appear the piecewise constant phenomenon.Experimental results show that our model gains 1~2dB higher PSNR than the class of zero-tree based on Bayes model threshold and P-M model.