一种改进的快速归一化互相关算法

An Improved Fast Normalized Cross Correlation Algorithm

摘要:

根据模板和全局最优子图的特点及其相互关系给出２个判据,对归一化互相关算法进行了改进.首先计算模板自相关值,再利用快速傅里叶变换方法计算互相关矩阵,利用第１个判据大幅缩小可能解的范围,减少匹配时间,然后利用第２个判据生成一个规模更小的候选最优解集合,最后确定全局最优解.实验结果说明,改进的归一化互相关算法能加快匹配速度,且能有效地提高图像匹配的准确率.

Abstract:

Some improvements were made for normalized cross correlation based on two criterions according to the characteristics of template and image and the interrelationship between them.The autocorrelation of the template was calculated at first,and the cross correlation between template and image was gained based on fast Fourier transform.Then the first criterion was used to shrink the range of the possible solutions, which could shorten the matching time;and the second criterion was applied to generating a more miniature set,in which the solution with the maximal normalized cross correlation was the global optimal solution.Experiments show that the normalized cross correlation based on the given criteria can speed the computing with an enhanced matching precision.