方差稳定变换下曲波域MR图像莱斯噪声去除算法  

MR images denoising for Rician noise using curvelet transform and variance stabilizing transformation

作　　者：孙中皋[1] 王巧玲 王新军[1] 王欣月 SUN Zhonggao;WANG Qiaoling;WANG Xinjun;WANG Xinyue(School of Physic and Electronic Technology,Liaoning Normal University,Dalian 116029,China)

机构地区：[1]辽宁师范大学物理与电子技术学院

出　　处：《辽宁师范大学学报（自然科学版）》2019年第3期335-343,共9页Journal of Liaoning Normal University:Natural Science Edition

基　　金：辽宁省教育厅科学研究青年项目(L201783643);辽宁省大学生创新创业训练计划项目(201710165000013)

摘　　要：磁共振图像往往含有莱斯噪声,不同于加性高斯噪声,莱斯噪声的分布与图像的数据相关,使其更为难以去除.已有方法表明,方差稳定变换可以将莱斯噪声分布变换为方差稳定的高斯分布.利用该特性,结合曲波变换这一新的多尺度变换理论,提出一种磁共振图像莱斯噪声去除算法.算法分别采用了硬阈值和贝叶斯软阈值两种曲波域去噪方法,并针对这两种方法对曲波域低频系数层未做有效去噪的缺陷,提出对低频系数层进行软阈值去噪的改进.实验结果表明,改进算法在峰值信噪比和平均结构相似度的评价上均有较大提高,证明该算法在去除莱斯噪声及保护图像信息上的有效性.Magnetic resonance image is often corrupted with Rician noise.Unlike additive Gaussian noise,Rician noise is signal-dependent which makes it difficult to separate signal from noise.Variance stabilizing transformation is developed to convert Rician distribution with variable variance to Gaussian distribution with constant variance.Utilizing this property,a new MR images denoising algorithm for Rician noise is proposed based on the new multi-scale transformation theory of curvelet transform.Two curvelet-domain denoising operations,hard thresholding and Bayesian soft thresholding,are used in the algorithm.Since these two methods do not denoise the curvelt coefficients of the low frequency subbands,this algorithm proposes a soft thresholding denoising method.Experimental results demonstrate that the proposed algorithm shows improved performance over the existing methods in terms of peak signal to noise ratio and structural similarity index measures.Comparative analysis reveals the efficacy of the proposed scheme to Rician noise removal while preserving the image details.

关 键 词：磁共振图像 莱斯噪声 曲波变换 方差稳定变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]