基于Zernike矩的快速图像配准算法

图像配准是机器视觉中的重要研究课题.本文针对任意角度图像配准问题,提出了一种基于Zemike矩的互相关算法.首先根据三角函数的对称性和反对称性,仅基于八分之一单位圆信息,快速计算Zemike矩；然后通过求取图像多个矩的互相关性,确定配准点位置；进而通过最小二乘拟合,得到亚像素配准位置；最终利用Zemike相位信息,估算旋转角度.同时,在线阶段采用径向多项式查表法加速配准过程.实验结果表明,该算法在保障配准精度的前提下显著提高了配准速度.

A Fast Image Registration Method Based on Zernike Moments

Image registration is an important research topic in machine vision. A cross correlation algorithm based on Zernike moments is developed to address the image registration problem with an arbitrary rotation angle. According to the symmetric and anti-symmetric properties of trigonometric functions, a fast approach is proposed to compute the Zernike moments, which gains the moments merely by calculation of 1/8 unit circle. Afterwards, the matching point is calculated by the correlations of multiple moments. Thereby, the sub-pixel matching position is achieved by Least Square Estimations (LSE). Meanwhile, the rotation angle is estimated by synthesizing the Zernike moments phase information. Finally, experimental results show that the proposed algorithm substantially improves the matching efficiency for a prescribed matching accuracy.