Image denoising has always been a hot issue in image processing. Therefore, hunting for a method of denoising efficiently and keeping the edge information simultaneously are goals people have been pursuing all the time. The basic idea of denoising is what is called average, so the main point is how to deal with smoothing while preserving details or high frequency parts. A. Buades, et al proposed the Nonlocal Means algorithm, which has changed the viewpoint of the problem.This thesis studies image denoising on wavelet domain and introduces the principles and algorithm of some denoising methods based on wavelet transform. The statistics of image wavelet coefficients is non-Gaussian and can be described by generalized Gaussian distribution (GGD). The thesis also investigates the issues of GGD statistical model for wavelet coefficients in a subband and the corresponding parameter estimation. The estimated parameters are used to define a generalized nonlocal means which allows us to restore the original image. A nonlocal means denoising algorithm on wavelet domain based on GGD statistical model for wavelet coefficients in all subbands is proposed. The simulation results indicate that our proposed method keeps a better visual result in edges information reservation and nosie removing. |