An improved MRI denoising algorithm based on wavelet shrinkage
Kaikai Song, Qiang Ling, Zhaohui Li, Feng Li · 2014
Magnetic resonance imaging (MRI) is very important in medical diagnosis. Denoising is a critical step for MRI diagnosis. Wavelet shrinkage is an efficient denoising method. It can be further classified into two types, the threshold method and the proportional-shrink method. However, both methods have their disadvantages. When the threshold method is implemented, the noise cannot be perfectly removed under a hard threshold while the denoised image may have fuzzy edges with a soft threshold. Furthermore, when the noise is too strong, the noise removal may not be enough by the threshold method. The proportional-shrink method requires that the variance field of the wavelet coefficients should change smoothly and the noise should obey a Gaussian distribution. If these assumptions are violated, the estimated ratios would not be precise so that too much texture information may be removed and the image can be distorted. This paper presents an improved method to combine the above two methods. By combining the processed results together, the improved method can achieve a good balance between denoising and retaining the texture information. We verify the efficiency of our method through some simulated data from an open database.