景象匹配相似性测度准则研究  

Classical similarity measures for scene matching

作　　者：安如[1] 王慧麟[2] 陶晓勋[1] 张琴[1] 陈春烨[1] 刘志辉[1] 

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]南京大学地理与海洋科学学院,江苏南京210093

出　　处：《河海大学学报（自然科学版）》2009年第2期147-152,共6页Journal of Hohai University(Natural Sciences)

基　　金：国家自然科学基金(40771137);中国科学院遥感应用研究所、北京师范大学遥感科学国家重点实验室开放基金;国家高科技研究发展计划(863计划)(2008AA12Z106)联合资助

摘　　要：对基于灰度的互信息、归一化积相关、序贯相似性检测和基于特征的点重复率、改进的Hausdorff距离以及AB匹配等几种典型的相似性测度方法进行了研究,探讨了这些方法在2种不同传感器、不同空间分辨率和不同季节拍摄的遥感景象匹配中的匹配成功率和匹配效率.试验表明,对于各种不同的地表景观,互信息、特征点重复率和改进的Hausdorff距离方法的匹配性能均好于归一化积相关以及序贯相似性检测算法,尤其当景象存在明显的灰度反转时,前者的匹配性能明显优于后者.对相似性测度准则的分析比较研究,有助于在实际应用中选择最合适的相似性测度匹配方法.Several typical similarity criteria for the intensity-based and feature-based matching methods were studied. They included the mutual information （MI）, the normalized product correlation （NPROD） and the sequential similarity detection algorithm （SSDA） methods for the intensity-based and the edge superposition ratio based on distance transformation （EDSRBDT）, the modified Hausdofff distance （MHD and M-HD） and the accumulated buffer （AB） methods for feature-based matching. Their success rates and time cost for matching were compared and discussed using remote sensing images with different spatial resolutions captured by two kinds of sensors in different seasons. The test results indicate that the matching performance of MI, EDSRBDT, MHD and M-HD is greatly superior to that of NPROD and SSDA for various land scenes, especially the images with obvious gray reversal. Comparative studies on the similarity criteria are useful for choosing the most reasonable similarity measures in practice.

关 键 词：相似性测度 景象匹配 图像匹配 有效性 匹配性能 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]