Abstract: Traditional image registration methods in frequency domain have two kinds: one is cross-correlation method; the other is phase-correlation method. The both methods obtain registration information by determining maximum peak in inverse Fourier Transforms domain. Cross-correlation method that compare with phase-correlation method has two disadvantages: one is that sometimes the presence of secondary maxima makes it difficult to select the actual maximum; moreover, the cross-correlation peak may be rather broad, so that the accuracy in determining its position may be poor. This paper proposes an improved cross-correlation method that is based on gradient pre-processing. Two pre-registering images are first pre-processed by gradient information before both images are registered with traditional cross-correlation. This method overcomes two disadvantages of traditional one and obtains precise results that have been proved by experiments.

Key words: image registration, cross-correlation, gradient, pre-processing.

摘要： 传统的频率域图像配准方法有两种：基于互相关的图像配准方法和基于相位相关的图像配准方法。这两种方法都是通过确定逆傅立叶变换域最大峰值位置来获得配准信息。互相关图像配准方法与相位相关图像配准方法相比，主要存在两个缺点：一是峰值的跨度过大，二是有时存在多个峰值。本文对传统方法进行了改进，提出了频率域基于梯度预处理互相关的图像配准方法，该方法首先对待配准图像进行梯度预处理，然后对预处理后的图像用传统的互相关方法进行配准。实验表明，该方法很好地克服了上述传统方法的不足，并能获得精确的配准结果。

关键词: 图像配准, 互相关, 梯度, 预处理