Multiwavelet-Based Estimation for Improving Magnetic Resonance Images

It is well known that magnetic resonance (MR) image data obey a Rician distribution, which introduces a signal-dependent biasto the data. Thinking over the Rician model, the text studies a translation-invariant (TI) multiwavelet thresholding technique and applies it to the squared magnitude MR images. Besides, a directional Wiener filter in the TI multiwavelet domain is presented using the adaptive neighborhood windows by the fact that the energy clusters in the three oriented wavelet subbands are different. Accordingly, the sizes of the windows should be adjusted as the scale changes. The optimum values were obtained by a large number of experiments. The denoising performance of the algorithm superior to other methods is demonstrated on both simulated and actual MR images.