Curvelet based image de-noising using beta-trim shrinkage for magnetic resonance images

This paper aims at the implementation of Curvelet transform for de-noising Magnetic Resonance images corrupted with Rician noise using a newly proposed technique called beta-trim shrinkage. In this paper beta-trim shrinkage is combined with Bayesian thresholding technique to recover the image corrupted with noise. The classical wavelet transform codes homogenous regions effectively. However for improved image perception edges need to be preserved. Curvelet transform is well suited for edge preservation. Curvelet transform offers a sharp detection of linear and curvilinear features thus providing visually high-resolution images. Experiments were performed on several images. Results show that a significant level of noise is reduced by the proposed beta-trim method using Bayes thresholding rule when compared to classical methods. An appreciably high value of Peak Signal to Noise Ratio (PSNR) and fairly lesser value of MSE (Mean square error) are also obtained by the proposed method.