显著图引导下基于偏互信息的医学图像配准  

Partial mutual information based medical image registration guided by saliency maps

作　　者：余时伟[1] 黄廷祝[1] 刘晓云[2] 陈武凡[2] 

机构地区：[1]电子科技大学数学科学学院,成都610054 [2]电子科技大学自动化工程学院,成都610054

出　　处：《仪器仪表学报》2013年第6期1218-1225,共8页Chinese Journal of Scientific Instrument

基　　金：973项目(2010CB732501);国家自然科学基金面上项目(61170311);四川省科技支撑重大项目(2012GZX0080)资助

摘　　要：提出了偏互信息(PMI)和归一化偏互信息(NPMI)相似性测度,PMI和NPMI分别是互信息(MI)和归一化互信息(NMI)的推广。建立了显著图引导图像配准(SMGR)框架,该框架利用显著图构建引导层进行图像粗配准,粗配准图像的几何拓扑特征与参考图像接近,并与参考图像构成新的待配准图像再进行精配准。对比实验表明,以偏互信息和归一化偏互信息为相似性测度,可以凸显图像的某些局部信息效用,利用提出的图像框架能提高医学图像的配准精度,并对噪声图像有较好的鲁棒性。同时,偏互信息和归一化偏互信息提供了图像处理中相似性测度的多种选择,SMGR框架方法丰富了图像配准方案。The partial mutual information （PMI） and normalized partial mutual information （NPMI） are proposed as two effective similarity measures. PMI and NPMI are the generalized forms of mutual information （MI） and normal- ized mutual information （NMI） , respectively. The saliency maps guided registration （SMGR） framework is estab- lished, and the framework uses saliency maps to establish guided layers and carry out image coarse registration. In sa- liency maps, the coarse aligned image is very close to the reference image, and the coarse registration image and refer- ence image are combined as a new image pair to be registered, on which the fine registration is performed. Contrast experiments demonstrate that using PMI and NPMI as similarity measures can highlight the effectiveness of some local information;using SMGR framework can improve the aligned accuracy of medical image and has good robustness to noisy image. Moreover, PMI and NPMI provide multiple selections for the similarity measure in image processing, and the method of SMGR framework enriches the schemes in image registration.

关 键 词：熵 互信息 偏互信息 显著图 配准 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]