基于差分多值的快速模板匹配算法  

Fast Template Matching Algorithm Based on Multi-Valued Differential

作　　者：吴小洪[1] 尧三品[1] 刘金保[1] 

机构地区：[1]广东工业大学机械装备制造及控制技术教育部重点实验室,广东广州510006

出　　处：《控制工程》2011年第3期349-351,共3页Control Engineering of China

基　　金：国家自然科学基金(50475044);教育部科技研究重点项目(2004106);广东省科技计划项目(2006A104D1003);广州市科技攻关项目(200623D9071)

摘　　要：基于归一化算法求解相似度原理,本文中提出了利用差分多值方法,在不降低匹配精度前提下大大的提高了匹配的速度。计算相似度时,先利用差分多值的方法对分子进行快速卷积运算,对其中一个式子进行差分,利用差分数组减少乘法运算;对分母利用邻窗口数据相关性进行运算,并对其中的一个式子用二次式展开,然后将累计项带入计算。在采用搜索策略时,综合比较遗传算法和金字塔式算法的优缺点基础上采用金字塔算法。在模板匹配过程中,取相似度最大的点作为匹配点。结果表明该算法匹配的速度快而又不降低匹配精度,而且稳定性好,具有一定的应用价值。Basing on normalized algorithm similarity principle, a multi- valued differential method is put forward to applying, greatly increasing speed of template matching under the premise of without reducing the accuracy of match. When calculating similarity, algo- rithm uses method of differential multiple - value for fast convolution operation for molecule, algorithm makes difference to one of statement holds; in order to reduce differential array multiplication, algorithm applies neighbor data relativity to calculate for denominator , making quadratic to one of statement holds and then into the calculation of total items. In the using of search strategies, comprehensively comparing advantages and disadvantages between genetic algorithm and pyramid algorithm, pyramid algorithm is chosen. During process of template matching, the thesis takes the biggest point as matching point. Results show that speed of algorithm is a little faster, without reducing matching accuracy and good stability, so algorithm has a certain and practical value.

关 键 词：差分运算 邻窗口数据 金字塔式搜索 模板匹配 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]