Research Of Image Denoising Algorithms Based On Wiener Filter

The Wiener filter is one of the classical linear filtering, and is known to be theoptimal estimator for the true underlying image. The Wiener filtering framework iswidely used in designing of various denoising algorithms. This thesis designs a few neweffective image denoising frameworks by deeply analyzing the Wiener filter. Theconcrete content includes five designs:(1) Design of Wiener filter in the gradientdomain by taking advantage of nonlinear diffusion;(2) Design of mixed functiondiffusion in gradient domain for image denoising to make the best use of differentdiffusion functions;(3) Design of image denoising method based on projection methodand Wiener filtering to overcome the deficiencies of the traditional wavelet imagedenoising algorithm, which does not make full use of the advantages of differentwavelets;(4) Design of multi-step local Wiener filter in wavelet domain and thestopping method rule based on noise variance in wavelet domain and gradient domain;(5) Design of a new wavelet threshold function from the point of view of parametertuning in the iterative diffusion. This new function used in the SURE-LET (Stein’sunbiased risk estimate and linear expansion of thresholds) method significantlyimproved the denoising ability. At the same time, we find that the traditional Wienerfilter is a single step discretization of modified continuous heat diffusion equation.Numerical experiments show that five algorithms designed in this thesis have commoncharacteristics of simple and efficient as well as excellent performance.