影像科学与光化学 ›› 2018, Vol. 36 ›› Issue (6): 539-550.DOI: 10.7517/issn.1674-0475.180905

• 论文 • 上一篇    

结合方差稳定变换和PGPD去除磁共振图像Rician噪声

张若男1, 池越1, 张伟2, 崔焘1   

  1. 1. 河北工业大学 电子信息工程学院, 天津 300401;
    2. 河北工业大学医院, 天津 300401
  • 收稿日期:2018-09-10 修回日期:2018-10-15 出版日期:2018-11-15 发布日期:2018-11-15
  • 通讯作者: 池越
  • 基金资助:
    国家自然科学基金(61401307)、河北省科学技术研究与发展项目(11213565)和河北省引进留学人员资助项目(CL201707)资助

Combining Variance Stable Transform and PGPD to Remove Rician Noise from Magnetic Resonance Images

ZHANG Ruonan1, CHI Yue1, ZHANG Wei2, CUI Tao1   

  1. 1. School of Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, P. R. China;
    2. Hebei University of Technology Hospital, Tianjin 300401, P. R. China
  • Received:2018-09-10 Revised:2018-10-15 Online:2018-11-15 Published:2018-11-15

摘要: 基于非局部自相似性块组学习的图像去噪(PGPD)算法去除高斯噪声效果优异,但是对于磁共振(MR)图像中Rician噪声的去除效果不理想。为此本文提出一种结合方差稳定变换和PGPD的新去噪算法FPGPD。该算法首先对含有Rician噪声的MR图像进行方差稳定变换,使噪声在变换域中近似服从高斯分布。用PGPD算法在变换域中去噪,最后经过方差稳定逆变换得到无偏去噪图像。理论分析和实验结果表明,FPGPD算法在去除MR图像中Rician噪声时比PGPD算法去噪性能好,具体体现为对图像细节和轮廓边缘保护得更好。

关键词: 磁共振图像, 去噪, 方差稳定变换, PGPD

Abstract: The patch group based non-local self-similarity prior learning for image denosing (PGPD) algorithm has excellent effect of removing Gaussian noise, but the removal effect of Rician noise in MR images is not ideal. In this paper, a new algorithm FPGPD combining variance stable transform and PGPD is proposed. Firstly, the algorithm performs the variance-stable transformation on the MR image with Rician noise, so that the noise approximation obeys the Gaussian distribution in the transform domain. Then, using the PGPD algorithm to denoise in the transform domain, and finally obtains the unbiased denoising image through the inverse variance transform. Theoretical analysis and experimental results indicate that the FPGPD algorithm has significantly improved denoising performance compared to the PGPD algorithm when removing Rician noise in MR images, which is better for image detail and contour edges.

Key words: MR image, digital image denoising, variance stable transformation, PGPD