基于互相关边界特性和图像积分的快速模板匹配算法  

Fast template matching algorithm based on boundary cross correlation and image integration

作　　者：吴小洪[1] 钟石明[1] 

机构地区：[1]广东工业大学机电工程学院,广州510006

出　　处：《计算机应用》2009年第7期1914-1917,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(50475044);教育部科技研究重点项目(2004106);广东省科技计划项目(2006A10401003);广州市科技攻关项目(2006Z3D9071)

摘　　要：基于归一化算法求解相似度原理,提出了综合利用互相关的边界条件和图像积分计算相似度的快速算法,在不降低匹配精度的前提下较大地提高了匹配速度。计算相似度时,归一化算法需要计算各位置的自相关值和互相关值,本算法先只计算自相关值,再利用H lder不等式原理,结合给定的边界阈值,剔除不满足条件的位置,减少其对应的互相关值计算。应用图像的积分进行匹配在于整个图像的积分可以在匹配之前进行计算,而在匹配过程中每一个子区域的自相关可以通过图像积分快速求得。本算法已在焊线机芯片识别系统中应用,结果表明该算法匹配的速度快而又不降低匹配精度,具有实际应用价值。A fast template matching algorithm was presented based on the theory of normalize cross correlation （NCC）. With the goal of accelerating the matching speed and yielding exactly the same result as other matching methods, it is applied to calculate the similarity of bounded cross correlation and image integration. In this process, NCC must calculate cross correlation and self-correlation at each point. Self-correlation was calculated only first and then Holder inequality was applied. With the given threshold, it can remove the unsatisfied point, and reduce the complex computation of cross correlation, significantly accelerate template matching. The advantage of image integration is in the process of matching the integration of sub-image can be obtained quickly by the integration of the whole image calculated before matching. The proposed algorithm has been applied in the wire-bonder image recognition system, and the experiment results show that template matching algorithm is of fast speed and high accuracy and this algorithm has practical value.

关 键 词：互相关边界特性 自相关 HLDER不等式 图像积分 模板匹配 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]