模板图像匹配中互相关的一种快速算法

基于归一化互相关系数的算法在模板匹配和特征跟踪中运用十分广泛,但缺点是其计算量很大.为此提出了一种在空间域利用盒形基简化互相关的快速算法,在不修改归一化互相关匹配原理的前提下,用原模板图像在一组正交盒形基张成的子空间上的投影取代原图像来进行互相关计算,以降低图像精度来缩减计算复杂度.实验说明,当搜索窗口大小较小时,此快速算法计算量明显小于传统的频域快速归一化互相关算法.

Fast Normalized Cross-Correlation for Template Matching

Algorithms based on normalized cross-correlation coefficients(NCC) are frequently used in template matching and feature tracking, but they have the drawback of high computational cost. By utilizing a set of box-like bases, we propose a new fast NCC computation algorithm taken in spatial domain. Without modifying the underline principle of NCC, we replace the original template image with its projection on subspace spanned by a set of box-like bases. This method can improve the computational efficiency though it will lose some precision. The given experiment shows that, when the size of the searching window is relatively small, our method requires significantly less computation than the traditional fast NCC algorithm in frequency domain.