基于位相相关检测的图像配准算法

Image registration algorithm based on phase-correlation detection

分析了传统的傅里叶-梅林变换和最近提出的虚拟极坐标傅里叶变换算法在户外图像注册中的局限性,提出了一种更加适合户外摄像机姿态注册的基于对数傅里叶变换的图像配准算法。该算法能够独立地计算沿两坐标轴方向的不同缩放因子和计算两图像间较大的旋转角度,扩展了基于傅里叶变换的图片配准算法的实际应用范围。在关键技术研究方面,提出了由升余弦函数构成的低通滤波器和一种快速收敛的角度搜索技术,使该算法达到了很高的效率和鲁棒性。

The defect of traditional Fourier-Mellin transformation and Pseudo Polar-FFT in outdoor image registration is analyzed. A novel image registration algorithm based on Log-EFT is developed, which is more suitable for calculating the cam- era posture and position in outdoor environment. This new algorithm can calculate large rotation and the scaling factors along X axis and along Y axis between two images independently, thereby the application field of image registration based on FFT is ex- tended. In the aspect of key techniques, a ccsine-window filter and a fast-converging technique of angular searching are proposed, the high efficiency and robustness are achieved using this new algorithm.