Abstract: Wavelet transformation has been widely used for image denoising. Different schemes of noise thresholding have been suggested. A good wavelet thresholding scheme should reflect the correct mapping of the wavelet coefficients of the noised images to that of clean images. In this study, we proposed a new method of image denoising, which utilizes neural networks to search the mapping relations in the wavelet domain. The method was applied to denoise MRI images and proven to be satisfactory in performance and insensitive to noisedistribution.


Key words: magnetic resonance imaging, neural network, wavelet transform