基于相位相关的亚像素配准技术及其在电子稳像中的应用  

Sub-pixel registration based on phase correlation and its application to electronic image stabilization

作　　者：孙辉[1] 李志强[1] 孙丽娜[1] 郎小龙[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《中国光学与应用光学》2010年第5期480-485,共6页Chinese Optics and Applied Optics Abstracts

基　　金：国家973重点基础研究发展规划资助项目(No.2009CB72400607)

摘　　要：介绍一种基于相位相关的图像配准方法,并将该方法应用于序列图像的运动检测,通过运动补偿,实现了序列图像的稳定输出。采用相位相关配准方法对序列图像进行亚像素精度配准,运用基于最小二乘的曲面拟合法,估计当前图像相对参考图像的亚像素级的平移量。在补偿图像运动时,为保证亚像素级的补偿精度,采用平滑算法,避免在图像补偿时出现马赛克现象。最后,对样本图像进行了亚像素级位移配准和图像补偿对比实验。实验结果表明,提出的方法可以检测到0.01pixel的运动量,最大配准误差为0.008pixel;采用亚像素级运动量补偿图像,最大误差<0.5。A novel method for the sub-pixel image registration based on phase correlation was introduced,which could be used to detect the movements of video sequences exactly,and could obtain the video outputs stably.Firstly,two images translated based on the phase correlation technique were registered,and the paraboloid surface fitting method based on least-square was used to estimate displacement parameters of the images.Then,the sub-pixel motion vector between the two sequential images could be exactly detected.A smooth algorithm in transform formula of the sub-pixel motion compensation was adopted to avoid the mosaics of images and to improve the precision of image compensation.In order to verify the algorithm in experiments,some sample simulations were performed.The results show that this method can detect the shift parameters accurately and efficiently,which makes it possible to estimate image displacement with 0.01-pixel accuracy.Further more,the maximum error of registration is only 0.008-pixel,and the root of mean square error of image compensation is also less than 0.5.

关 键 词：相位相关 亚像素 图像配准 电子稳像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]