自适应连续多级分区与初始阈值估计的快速模板匹配方法  

Fast template matching algorithm based on AMSP and initial threshold estimation

作　　者：汪鲁才[1] 易锡年[1,2] 陈小天[3] 刘鑫[1] 

机构地区：[1]湖南师范大学工学院,湖南长沙410081 [2]湖南三一重工股份有限公司,湖南长沙410073 [3]国防科学技术大学ATR国防重点实验室,湖南长沙410073

出　　处：《红外与激光工程》2013年第4期1106-1111,共6页Infrared and Laser Engineering

基　　金：湖南省科技厅科研基金(2010FJ6014)

摘　　要：归一化互相关测度在光照改变时比采用绝对差之和测度(SAD)要稳定,但是归一化互相关测度的缺陷在于它的计算量非常大。为此,提出了一种结合自适应连续多级分区和初始阈值估计的基于归一化互相关(NCC)的快速模板匹配算法。根据模板图像中不同模块的梯度值,将模板图像进行逐级分区,通过分区顺序将互相关之和分为不同的层,得到各层互相关的上界,运用柯西-施瓦兹不等式得到上界间的关系,形成自适应连续多级分区淘汰方法。同时,为了加快匹配速度,利用初始阈值估计产生一个较大的边界阈值,以淘汰初始搜索时的大量非匹配点,减少搜索点数目。实验结果表明:所提出的算法具有较好的鲁棒性,且算法的执行速度优于传统算法。The Normalization Cross Correlation （NCC） me asure is more stable than Sum of Absolute Differences （SAD） measure when the illumination changes. However, it needs large calculated amount, which is its disadvantage. Therefore, a fast template matching algorithm based on NCC combing Adaptive Multilevel Successive Partitioning （AMSP） with the initial threshold estimation was proposed in this paper. The template image was partitioned into different blocks steeply according to the gradient values of the different modules in the template image, the summation of cross correlation was partitioned into different levels with the partition order to get the upper bounds of each layer, and the Cauchy-Schwartz inequality was used to get the relation between different upper bounds, then the approach of adaptive multilevel successive partitioning elimination was formed. In order to further accelerate the matching speed, the initial threshold estimation was used to generate a large boundary threshold, which could eliminate lots of unmatched points as initial searching and reduce the number of search points. The experimental results demonstrate that the proposed algorithm has strong robustness, and the execution speed of the proposed approach is superior to traditional algorithms.

关 键 词：快速模板匹配 自适应连续多级分区 归一化互相关 部分边界相关 初始阈值估计 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]