规则图像特征的标准互相关相似函数  

Normalized cross correlation computation for geometry image features

作　　者：孙明磊[1] 张融[1] 朱小峰[2] 宗光华[1] 

机构地区：[1]北京航空航天大学机械工程及自动化学院,北京100191 [2]北京市劳动保护科学研究所,北京100054

出　　处：《北京航空航天大学学报》2008年第12期1441-1444,共4页Journal of Beijing University of Aeronautics and Astronautics

基　　金：中国博士后科学基金资助项目(20070420287);国家863计划资助项目(2004AA404260)

摘　　要：利用标准互相关算子对机器视觉领域常用的十字图像特征进行匹配运算,发现相似函数在峰值区域的分布特征为4个互相对称的双曲面.受此启发,对规则几何图像特征(如矩型、圆型等)的标准互相关相似函数展开了研究.以图像特征概率分布的形式给出了NCCO算子在二值图像匹配下的相似函数计算式.推导并证明了矩型、圆型等典型图像特征相似函数在峰值点区域具有确定的数学模型.基于上述数学模型,在微装配显微视觉系统上进行了实验验证,对模板优化、图像特征的尺寸等优化方法进行了探讨.上述结论对图像定位、图像标定、图像跟踪等技术研究具有实用价值.Normalized cross correlation operator （NCCO） was used for pattern matching to localize popular cross-shaped features in microscopic vision. Distribution feature of similarity function around peak zone was found to be four symmetry hyperboloid planes. Inspired by this, some research work on pattern matching of geometry image features （such as rectangular, circular, etc. ） was presented. A probability distribution based formula computing NCCO with two binary images was proposed. Mathematic models of some typical geometry image features＇ （rectangular, circular, cross-type, box-type） similar functions around peak point were derived from and proved. Based on these models, some experiments about template image optimum and feature size optimum were conducted on a microscopic vision workcell. Conclusions above are practically useful to image positioning, image calibration and image tracking techniques.

关 键 词：视觉 图像 相关方法 模式匹配 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置]