一种改进的病理显微图像亚像素快速配准方法
|
摘要
针对病理显微图像的快速高精度配准的应用需求,提出了一种改进的亚像素快速配准方法。通过采用对数差值函数对显微图像重叠区域进行信息量评估,由H型对数差值模板配准获取两幅图像像素级的粗定位和有效子图,继而在两幅子图上采用局部上采样相位相关法来获取亚像素级别的配准估计值;该方法利用H型对数差值模板简洁、配准速度快的优点规避了相位相关法计算量随着图像尺寸增大而急剧增长的缺陷,并能够对显微图像进行信息量评估以避免空白图像造成的误配;经实验表明,改进的显微图像配准方法配准精度可达0.01像素,速度为相位相关法的3.7倍(图像大小为2448×2048),更适合应用于病理显微图像的快速高精度配准。
Aiming at the application requirements of fast and high-precision registration of pathological microscopic images,an improved sub-pixel fast registration method is proposed.By using the logarithmic difference function to evaluate the overlap region of the microscopic image,the coarse positioning and effective subgraph of the two images are obtained by H-type logarithmic difference template registration.Then local up-sampling phase correlation is used on the two subgraphs to obtain sub-pixel level registration.The method takes the advantages of fast registration speed of H-type logarithmic difference template to avoid the defect that the calculation of phase correlation method increases sharply with the increase of image size.Besides,the method avoid mismatch caused by the blank image.Experiments show that the improved microscopic image registration method has a registration accuracy of0.01 pixels and a speed of 3.7 times that of the phase correlation method(image size is 2448×2048),which is more suitable for fast and high-precision matching of pathological microscopic images.
Aiming at the application requirements of fast and high-precision registration of pathological microscopic images,an improved sub-pixel fast registration method is proposed.By using the logarithmic difference function to evaluate the overlap region of the microscopic image,the coarse positioning and effective subgraph of the two images are obtained by H-type logarithmic difference template registration.Then local up-sampling phase correlation is used on the two subgraphs to obtain sub-pixel level registration.The method takes the advantages of fast registration speed of H-type logarithmic difference template to avoid the defect that the calculation of phase correlation method increases sharply with the increase of image size.Besides,the method avoid mismatch caused by the blank image.Experiments show that the improved microscopic image registration method has a registration accuracy of0.01 pixels and a speed of 3.7 times that of the phase correlation method(image size is 2448×2048),which is more suitable for fast and high-precision matching of pathological microscopic images.
引文
[1]Ferrante E,Paragios N.Slice-to-volume medical image registration:a survey[J].M-edical Image Analysis,2017:S1361841517300701.
[2]Paul S,Pati U C.Remote Sensing Opt-ical Image Registration Using Modified U-niform Robust SIFT[J].IEEE Geoscience and Remote Sensing Letters,2016,13(9):1300-1304.
[3]Fung K M,Hassell L A,Talbert M L,et al.Whole slide images and digital m-edia in pathology education,testing,and practice:the Oklahoma experience.[J].An-alytical Cellular Pathology,2016,35(1):37-40.
[4]Clunie D,Hosseinzadeh D,Wintell M,et al.Digital Imaging and Communications in Medicine Whole Slide Imaging Connectathon at Digital Pathology Association Pathology Visions 2017[J].Journal of Pathology Informatics,2018,9(1):6.
[5]Fred R.Dee.Virtual microscopy in pathol-ogy education[J].Human Pathology,2009,40(8):1112-1121
[6]Appleton B C,Bradley A P,Wilderm-oth M.Towards Optimal Image Stitchi-ng for Virtual Microscopy[A].Digital I-mage Computing:Techniques&Applic-ations[C].IEEE,2005.
[7]彭勋.数字化显微镜全切片扫描关键技术研究[D].上海:上海交通大学,2014.
[8]Phase correlation imaging of unlabeled ce-ll dynamics:[J].Scientific Reports,2016,6:32702.
[9]余应淮,谢仕义,梅其祥.基于核回归修正的上采样相位相关精确运动估计算法[J].计算机应用,2016,36(8):2316-2321.
[10]马宁,门宇博,门朝光,et al.基于扩展相位相关的小基高比立体匹配方法[J].电子学报,2017(8):1827-1835.
[11]Smith Julius O.Mathematics of the discret-e.Fouriertransform[M].Stanford.CCRMA,2002:145-179
[12]Soummer R,Sivaramakrishnan A,P-ueyo L,et al.Fast computation of Ly-ot-style coronagraph propagation[J].Op-tics Express,2007,15(24):15935-15951.
[13]邓严.医学显微图像配准及存储技术研究[D].武汉:华中科技大学,2013.
[2]Paul S,Pati U C.Remote Sensing Opt-ical Image Registration Using Modified U-niform Robust SIFT[J].IEEE Geoscience and Remote Sensing Letters,2016,13(9):1300-1304.
[3]Fung K M,Hassell L A,Talbert M L,et al.Whole slide images and digital m-edia in pathology education,testing,and practice:the Oklahoma experience.[J].An-alytical Cellular Pathology,2016,35(1):37-40.
[4]Clunie D,Hosseinzadeh D,Wintell M,et al.Digital Imaging and Communications in Medicine Whole Slide Imaging Connectathon at Digital Pathology Association Pathology Visions 2017[J].Journal of Pathology Informatics,2018,9(1):6.
[5]Fred R.Dee.Virtual microscopy in pathol-ogy education[J].Human Pathology,2009,40(8):1112-1121
[6]Appleton B C,Bradley A P,Wilderm-oth M.Towards Optimal Image Stitchi-ng for Virtual Microscopy[A].Digital I-mage Computing:Techniques&Applic-ations[C].IEEE,2005.
[7]彭勋.数字化显微镜全切片扫描关键技术研究[D].上海:上海交通大学,2014.
[8]Phase correlation imaging of unlabeled ce-ll dynamics:[J].Scientific Reports,2016,6:32702.
[9]余应淮,谢仕义,梅其祥.基于核回归修正的上采样相位相关精确运动估计算法[J].计算机应用,2016,36(8):2316-2321.
[10]马宁,门宇博,门朝光,et al.基于扩展相位相关的小基高比立体匹配方法[J].电子学报,2017(8):1827-1835.
[11]Smith Julius O.Mathematics of the discret-e.Fouriertransform[M].Stanford.CCRMA,2002:145-179
[12]Soummer R,Sivaramakrishnan A,P-ueyo L,et al.Fast computation of Ly-ot-style coronagraph propagation[J].Op-tics Express,2007,15(24):15935-15951.
[13]邓严.医学显微图像配准及存储技术研究[D].武汉:华中科技大学,2013.