基于图像特征和光流场的非刚性图像配准

发布时间：2018-03-05 22:09

本文选题：图像配准　切入点：非刚性配准　出处：《光学精密工程》2017年09期 　论文类型：期刊论文

【摘要】：考虑传统非刚性图像配准方法无法同时满足配准精度和配准时间要求,综合图像的特征和灰度信息,提出了几种改进的非刚性图像配准方法:基于圆形描述子特征的非刚性配准方法(Circle Descriptor Feature,CDF),基于动态驱动力Demons的非刚性配准方法(Dynamic Driving Force Demons,DDFD),和基于图像特征和光流场的非刚性配准方法。CDF方法通过提取图像的特征点,采用圆形描述子代替传统方法的正方形描述子来保证图像的旋转不变性,提高配准速度;DDFD方法通过引入驱动力系数动态改变驱动力,有效地解决了传统方法配准时间和配准精度低的问题;基于图像特征和光流场的非刚性配准方法则首先提取浮动图像和参考图像的特征点,然后利用提取的特征点进行粗配准(特征级配准),再采用基于光流场的方法进行精细配准(像素级配准),最终实现配准精度和配准时间的兼顾。对checkboard测试图像、自然图像、脑部MR图像、肝部CT图像进行了实验测试,结果表明,本文方法在配准时间、配准精度及对大形变图像的适应性方面均优于传统尺度不变特征转换(SIFT)、加速鲁棒特征(SURF)、Demons、Active Demons和全变差正则项-L~1范数项(TV-L~1)等方法。
[Abstract]:Considering that the traditional non-rigid image registration method can not meet the registration accuracy and registration time requirements at the same time, the image features and gray level information are synthesized. Several improved non-rigid image registration methods are proposed: circle Descriptor feature based non-rigid registration method based on circular descriptor feature, dynamic Driving Force Demonsd FDF based on dynamic driving force Demons, and image feature and optical flow field. The non-rigid registration method. CDF method extracts the feature points of the image. The circular descriptor is used to replace the square descriptor of the traditional method to ensure the rotation invariance of the image, and the dynamic driving force is changed by introducing the driving force coefficient to improve the registration speed. It effectively solves the problems of low registration time and registration accuracy in traditional methods, and the non-rigid registration method based on image feature and optical flow field firstly extracts the feature points of floating image and reference image. Then the extracted feature points are used for rough registration (feature gradation), and then fine registration (pixel gradation) based on optical flow field is used to achieve both registration accuracy and registration time. Mr images of brain and CT images of liver were tested experimentally. The results showed that the registration time of this method, The registration accuracy and adaptability to large deformation images are better than those of the traditional scale invariant feature conversion (SIFT), and the methods of accelerating robust features such as robust Demons and TV-L ~ (1)) are also discussed in this paper, and the results are as follows: (1) the accuracy of registration and the adaptability to large deformation images are better than those of the traditional scale-invariant feature conversion (SIFT).
【作者单位】：山东大学(威海)机电与信息工程学院;
【基金】：国家自然科学基金资助项目(No.81671848,No.81371635) 山东省重点研发计划资助项目(No:2016GGX101017)
【分类号】：TP391.41

【相似文献】