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。
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.
出处
《计算机应用研究》
CSCD
北大核心
2006年第11期1-3,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(60234010
60434020)
国家"973"计划资助项目(2002CB312200)
关键词
相似性度量
匹配精度
捕获/匹配/虚警概率
Similarity Measure
Registration Precision
Capture/Registration/False-alarm Probability
