基于融合式PC-ORB的异源图像配准算法  

Heterogeneous Image Registration Algorithm Based on Fusion PC-ORB

作　　者：伍朗 易诗 陈梦婷 李立 WU Lang;YI Shi;CHEN Mengting;LI Li(School of Mechanical and Electrical Engineering,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]成都理工大学机电工程学院,四川成都610059

出　　处：《红外技术》2024年第4期419-426,共8页Infrared Technology

基　　金：四川省科技厅重点研发项目(2021YFGO075,2021YFGO076);四川省车辆测控与安全重点实验室开放基金(OCCK2021-008);四川省重点科技项目(2020ZDZX0019);成都理工大学2021—2023年高等教育人才培养质量和教学改革项目(JG2130109,JG2130216)。

摘　　要：异源图像配准中,由于图像的成像机理差异,图像像素强度关联和旋转畸变是不可避免的两大问题,针对图像像素强度关联问题,提出了基于辐射不变特征变换(radiation-variation insensitive feature transform,RIFT)的图像配准算法,对图像间像素关联差异小的图像对配准有良好的精度,但对旋转畸变图像会产生较多错误匹配。对于旋转畸变问题,传统的ORB(oriented fast and rotated brief)算法,对旋转图像的配准有一定的稳定性,但对于强度变化不明显的图像对,特征点检测质量较低,配准精度不理想。因此本文将相位一致性(phase consistency,PC)融合进ORB算法,利用相位信息代替传统的图像强度信息,再构造旋转不变性BRIEF特征描述子,对图像像素强度变化和旋转畸变均具有鲁棒性。用图像像素强度关联不明显的红外图像和可见光图像进行配准实验,本文算法针对不同旋转幅度的图像的配准精度较高,RMSE稳定在1.7~2.1,优于RIFT算法,在特征点检测数量、配准精度和效率等性能上均有良好性能。In heterogeneous image registration,because of the differences in the imaging mechanisms,image pixel intensity correlation and rotation distortion are two inevitable problems.Aiming at the problem of image pixel intensity correlation,an image registration algorithm based on a radiation-invariant feature transform(RIFT)is proposed;it has good accuracy for image registration with small differences in the pixel correlation between images,but produces more error matching for rotation distortion images.For the problem of rotational distortion,the traditional Oriented Fast and Rotated Brief(ORB)algorithm has a certain degree of stability in the registration of rotating images;however,for image pairs with insignificant intensity changes,the quality of the feature point detection is low and the registration accuracy is not ideal.Therefore,this study integrates Phase Consistency into the ORB algorithm,replaces traditional image strength information with phase information,and constructs a rotation-invariant BRIEF feature descriptor that is robust to changes in the pixel strength and rotation distortion in the image.The registration experiment is conducted using infrared and visible-light images with unclear pixel intensity correlations.The algorithm proposed in this paper has high registration accuracy for images with different rotation amplitudes,and the RMSE is stable at 1.7-2.1,which is superior to the RIFT algorithm.It performs well in detecting a large number of feature points,achieving high registration accuracy,and maintaining efficiency.

关 键 词：图像配准 特征匹配 相位一致性 旋转不变性 ORB算法 

分 类 号：TN911.73[电子电信—通信与信息系统]