Research Of Wavelet Image Denoising And Its Application On MR Images

Wavelet image denoising is an important method of image denoising. It can remove the Gaussian noise more efficiently than the traditional methods. Recently, many different schemes of wavelet image denoising were proposed. In this thesis, the theory of wavelet transformation is introduced and several classical wavelet image denoising schemes are described and analyzed. Based on the NeighShrink scheme suggested by G. Y. Chen et. al., which has been proved very efficient, we proposed two new wavelet image denoising schemes, namely Enhanced NS (ENS) and Apdative NS respectively. The former can be used to enhance image details while denoising the image while the latter can adjust its parameters automatically according to the amount of the details in images, thus reserves more details while denoising.Because of the effectiveness and the advantages of the wavelet denoising, many people have applied it on MR image and achieved good results. However, there remain some problems in the existing algorithms. For example, some of them were developed assuming the noise in magnitude MR images to be additive Gaussian which is not correct in fact, and some of them were proposed to denoise the real and the imaginary parts of signal independently, which can bring phase distortions. While some other algorithms may be void of the mentioned problems, their ways of estimating the noise need to be improved. Based on the study of the distribution of the noise in MR images, we proposed a new method to estimate the noise in MR image. Additionally, based on NeighShrink and another scheme proposed by Nowak, we bring out a new scheme to denoise the squared magnitude MR image which estimates the additive noise and the signal correlated noise separately. Efficiency of this new method has been proved by experiments.