首页 | 本学科首页   官方微博 | 高级检索  
     

几种经典相似性度量的比较研究
引用本文:刘宝生,闫莉萍,周东华. 几种经典相似性度量的比较研究[J]. 计算机应用研究, 2006, 23(11): 1-3
作者姓名:刘宝生  闫莉萍  周东华
作者单位:清华大学,自动化系,北京,100084;空军装备研究院,北京,100085;清华大学,自动化系,北京,100084
基金项目:国家自然科学基金;国家重点基础研究发展计划(973计划)
摘    要:以图像匹配概率和匹配精度为准则,对几种经典的相似性度量进行了比较研究。得出绝对差(AD)与平均绝对差(MAD)算法等价;平方差(SD)、平均平方差(MSD)、积相关(Prod)与归一化积相关(Nprod)算法等价的结论。当图像信噪比较低时,Prod/Nprod/SD/MSD算法优于AD/MAD算法;否则,后者优于前者。相似性度量的比较研究可为实际应用中图像配准或景象匹配问题中相似性度量的选择提供重要的参考依据。

关 键 词:相似性度量  匹配精度  捕获  匹配  虚警概率
文章编号:1001-3695(2006)11-0001-03
收稿时间:2005-09-20
修稿时间:2006-03-30

Comparison of Some Classical Similarity Measures
LIU Bao sheng,YAN Li ping,ZHOU Dong hua. Comparison of Some Classical Similarity Measures[J]. Application Research of Computers, 2006, 23(11): 1-3
Authors:LIU Bao sheng  YAN Li ping  ZHOU Dong hua
Affiliation:(1.Dept. of Automation, Tsinghua University, Beijing 100084, China;2.Equipment Academy of Airforce, Beijing 100085, China)
Abstract:In view of the image registration probability and registration precision,some classical similarity measures are compared theoretically.The conclusions drawn from this paper are: Absolute Difference(AD) and Mean Absolute Difference(MAD) are equivalent.Square Difference(SD),Mean Square Difference(MSD),Product correlation similarity measure(Prod) and the Normalized product correlation similarity measure(Nprod) are equivalent.Prod/Nprod/SD/MSD is better than AD/MAD when Signal-Noise-Ratio(SNR) is low,otherwise AD/MAD is better than Prod/Nprod/SD/MSD.The comparison of the classical similarity measures may be used as a reference for the choice of similarity measures when some practical image registration or scene matching problems are encountered.
Keywords:Similarity Measure  Registration Precision  Capture/Registration/False-alarm Probability
本文献已被 CNKI 维普 万方数据 等数据库收录!
点击此处可从《计算机应用研究》浏览原始摘要信息
点击此处可从《计算机应用研究》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号