影像特征匹配算法性能评价研究  

Performance evaluation of the image feature matching algorithm

作　　者：孙晓昱 王子阳 张楠 古林玉[4] SUN Xiaoyu;WANG Ziyang;ZHANG Nan;GU Linyu(Tianhui Satellite Center of China,Beijing 100094,China;Unit 31682,Lanzhou 730000,China;Unit 61206,Beijing 100042,China;Information Engineering University,Zhengzhou 450000,China)

机构地区：[1]中国天绘卫星中心,北京100094 [2]31682部队,甘肃兰州730000 [3]61206部队,北京100042 [4]信息工程大学,河南郑州450000

出　　处：《测绘科学与工程》2020年第4期31-39,共9页Geomatics Science and Engineering

摘　　要：现阶段对不同影像特征匹配方法性能的评价,主要聚焦方法在描述过程中具备的各类不变性,缺乏在各类实际应用需求下的定量描述。本文对当前主流的13种影像特征匹配方法进行分类梳理,设计了面向应用的影像特征匹配方法评价指标,利用公开数据集,从鲁棒性、准确度、充分性和效率四个方面评价了各方法在不同应用场景下的性能,为匹配方法适配典型应用场景提供思路参考。The current evaluation of image feature matching algorithms mainly focuses on the invariability in the description process,which lacks quantitative evaluation in different applications.In this paper,13 classic matching algorithms are classified and sorted,and then the evaluation indexes for image feature matching method oriented to application are designed.Based on the open data set,the performance of each method in different applications is evaluated in terms of robustness,accuracy,sufficiency and efficiency.The experimental results show that the proposed evaluation indexes can provide a reference for choosing correct matching method in typical application scenarios.

关 键 词：特征匹配 评价指标 鲁棒性 准确度 充分性 效率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]