摘要
针对影像视角、光照或尺度变化较大,目前常用特征提取算法匹配效果不理想的问题,提出一种用于联合检测特征点和提取特征描述子的多尺度网络,在特征点检测和描述中融入多尺度信息,提升特征点的鲁棒性。通过在影像数据上对该算法进行实验验证,发现该算法可以有效提高特征点的可重复性以及特征描述子的辨识力,从而提升匹配精度,并且可以得到更准确的单应估计结果。
In order to solve the problem that current commonly used feature point extraction algorithms is not ideal when viewpoint,illumination or scale change a lot,a multi-scale network is proposed for joint feature point detection and description.Multi-scale information is integrated into feature point detection and description to improve the robustness of feature points.Experimental results show that the proposed algorithm can effectively improve the repeatability of feature points and the discriminating power of feature descriptors,thus improving the matching accuracy and obtaining more accurate homography estimation results.
作者
王颖
龚烨
尹泓澈
李礼
姚剑
WANG Ying;GONG Ye;YIN Hongche;LI Li;YAO Jian(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;AI Application and Innovation Research Center,The Open University of Guangdong,Guangzhou 510091,China)
出处
《测绘地理信息》
CSCD
2022年第S01期167-171,共5页
Journal of Geomatics
关键词
多尺度
影像匹配
特征点检测
特征描述
multi-scale
image matching
feature point detection
feature point description
