基于SURF特征与边缘信息的图像配准  

Image Registration Based on SURF Features and Edge Information

作　　者：刘波 赵于前[1] 刘彬旭[1] 

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083 [2]湖南省财贸医院,湖南长沙410002

出　　处：《中国医学物理学杂志》2011年第6期3000-3003,3024,共5页Chinese Journal of Medical Physics

基　　金：国家自然科学基金资助项目(No.61172184);中国博士后科学基金特别资助项目(No.200902482)

摘　　要：目的:针对特征图像配准方法速度快、效率高,而医学图像具有结构信息不明显、变形复杂等特点,该类方法常常失效,提出一种结合SURF特征与图像边缘信息的配准方案。方法:首先使用SURF法检测图像的特征点,采用角度法判断特征点是否匹配错误,并将误匹配的特征点删除;然后,将所得特征点对与图像的Canny边缘相结合,形成一个新的特征点集;最后使用TPS-RPM法对该特征点集进行配准。结果:采用边缘点与SURF特征结合的配准方案,减少了特征点误匹配产生的不良影响,提出的筛选特征点方法能有效地删除部分匹配错误的特征点对。结论:该方案比只用SURF特征匹配的结果更为精确。Objective: For the fast and high efficiency of image registration methods for features,they often fail in medical image registration because the structural information of medical images are not obvious and the deformation of medical images is always complex,a method combined SURF features and edge information is presented for image registration.Methods: First,the SURF is used for features detection.Then the mismatched points are found and deleted according to the angles between features.Next,the obtained feature point couples are combined with Canny edge to construct a new set for feature points.Finally,the TPS-RPM method is used to conduct image registration for the feature points set.Results: The image registration method that combines the edge points and the SURF features can reduce the error caused by feature points mismatching,and the proposed feature points selecting method can partly delete the mismatched feature point couples effectively.Conclusions: The proposed image registration method is more effective than the SURF image registration method.

关 键 词：图像配准 SURF Canny边缘 

分 类 号：R312[医药卫生—基础医学]