WCSE

MR Image Denoising Algorithm Based on Variance Stable Transformation and LSM-NLR

Abstract—The denoising algorithm based on Laplacian scale mixing model and non-local low-rank approximation (LSM-NLR) has a very good denoising effect for mixed gaussian white noise images, but the suppression effect is not ideal for spatial variation Rician noise in MR images with low signal-to-noise ratio. Therefore, this paper proposes a new denoising method combining variance stable transformation and LSM-NLR. This method has used the noise estimated by modified median absolute deviation to carry out the variance stable transformation of the noise image amplitude, so that the noise is independent of the signal amplitude and spatial position. LSM-NLR algorithm can be used to suppress the noise of transformed image, and finally obtain the unbiased denoised image through the variance stable inverse transformation. Experimental results have demonstrated that the proposed algorithm can suppress MR image noise effectively and protect details of image. The value of PSNR and SSIM were slightly higher in the simulation experiment. The NIQE value and BIQI value of the proposed algorithm were slightly lower and higher from the de-noising results of Real breast MR images

Index Terms— Rician noise, variance-stabilization transformation, LSM, Nonlocal Low-rank

Liling Yu, Binfeng Xu, Gangping Zhang, Wenping Liu, Haijun Chen, Jingyu Guo
Guangdong Food and Drug Vocational College, CHINA

[Download]

Cite: Liling Yu, Binfeng Xu, Gangping Zhang, Wenping Liu, Haijun Chen, Jingyu Guo , " MR Image Denoising Algorithm Based on Variance Stable Transformation and LSM-NLR " Proceedings of 2020 the 10th International Workshop on Computer Science and Engineering (WCSE 2020), pp. 179-184, Shanghai, China, 19-21 June, 2020.