Improvement of residual noise characteristics in FREBAS MR image denoising

The purpose of this study was to improve the residual noise characteristics in FREBAS image denoising. We have proposed the FREBAS transform as a new multi-resolution image analysis method and have applied this method to MR image denoising. However, when FREBAS denoising is applied to images with a low signal-to-noise ratio (SNR), prominent noise showing a characteristic pattern is observed. To reduce such noise, a new FREBAS denoising method is proposed. The denoising steps are Fresnel transform of the noisy input images, the application of FREBAS denoising to the Fresnel-transformed images, and finally, inverse Fresnel transform of the denoised images. The characteristic noise appearing in the Fresnel-transformed signal is scrambled by the inverse Fresnel transform in the final step. Simulations and experimental studies showed that the prominent noise is dramatically reduced and that the residual noise characteristics are improved. The increase in SNR is almost the same and image degradation is somewhat improved compared with the conventional FREBAS denoising method. Comparison with other denoising filters demonstrated the excellent denoising performance of the proposed method. (author)