An Improved Adaptive Wiener Filtering Algorithm

In this paper, an anisotropic image denoising algorithm is proposed by combining a nonlinear version of the local structure tensor together with Wiener filtering, where the shape and size of smoothing windows are determined by an iteratively updated nonlinear diffusion process while those in the Wiener filter are fixed. In this way, the method is data-adaptive and helps to better preserve boundaries and reduce structure delocalization. An additive operator splitting scheme is applied to solving nonlinear diffusion equation to improve computational efficiency. In simulations, the approach exhibits better performance and significant peak signal-to-noise ratio improvement than Wiener filtering and some wavelet-based filtering schemes, particularly in edge regions