基于抽样随机一致性和图像特征的图像配准  

Random sampling consensus-based and feature-based image registration

作　　者：段嘉旭 罗林 高晓蓉 彭建平 李金龙 

机构地区：[1]西南交通大学物理科学与技术学院光电工程研究所,成都610031

出　　处：《信息技术》2017年第11期100-104,共5页Information Technology

摘　　要：文中介绍了一种基于Harris角点检测图像特征的图像配准方法。通过随机一致性算法(RANSAC)和正则化交叉相关测度建立了待配准图像和模板图像之间特征点的对应关系。分别采用仿射变换和透射变换通过该对应关系进行图像配准。在完成配准后,分别采用自适应B样条插值算法和多项式插值算法对配准后的图像进行校正。采用信噪比和交叉相关系数作为评价标准,通过对几种方法相结合的图像处理结果进行比较,得出结论:对应超声TOFD图像,仿射变换结合B样条插值校正的方法可以得到最好的效果。A Harris corner detection based image registration algorithm is presented in this paper. The correspondence between the module and the sensed image is established via random sampling consensus（ RANSAC） with normalized cross-correlation（ NCC）. After image registration,adaptive B-spline and polynomial is adopted for calibration. SNR and cross-correlation is taken into consideration as an evaluation standard. It comes to： in the process of ultrasonic TOFD images,affine transformation with Bspline interpolation can achieve a better result.

关 键 词：图像配准 RANSAC 对应关系 NCC B样条插值 多小时插值 信噪比 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]