基于改进SIFT和SURF算法的沙丘图像配准  

Registration of sand dune images using an improved SIFT and SURF algorithm

作　　者：唐颖复 王忠静[1,3,4] 张子雄 TANG Yingfu;WANG Zhongjing;ZHANG Zixiong(Department of Hydraulic Engineering,Tsinghua University,Beijing 100084,China;China Institute of Water Resources and Hydropower Research,Beijing 100038,China;State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China;State Key Laboratory of Plateau Ecology and Agriculture,Qinghai University,Xining 810016,China)

机构地区：[1]清华大学水利水电工程系,北京100084 [2]中国水利水电科学研究院,北京100038 [3]清华大学水沙科学与水利水电工程国家重点实验室,北京100084 [4]青海大学省部共建三江源生态与高原农牧业国家重点实验室,西宁810016

出　　处：《清华大学学报（自然科学版）》2021年第2期161-169,共9页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金项目(41671020);国家重点研发计划项目(2016YFC0402900);宁夏重点研发计划项目(2020BCF01002)。

摘　　要：单个沙丘的图像配准受到沙丘图像颜色相近、纹理相似和轮廓模糊等问题困扰,常用的特征提取和特征点配准方法易产生较多的错误匹配点。为了实现有效的单个沙丘跟踪,该文提出了适用于沙丘图像的基于相似三角形原理的尺度不变特征变换(SIFT)和快速鲁棒特征(SURF)的特征点筛选算法。该算法利用暴力匹配法匹配SIFT与SURF特征点,先根据K最近邻算法(KNN)初步筛选匹配点,再利用相似三角形原理对匹配点进一步筛选。将该算法应用于库姆塔格沙漠的沙丘图像配准,实例表明所提出的算法是一种适用性强、准确率高的沙丘图像配准算法。The image registration of a single sand dune is difficult due to the similar colors,similar textures and blurred edges of the dunes.Commonly used methods such as feature extraction and feature point matching tend to have mismatched points.This paper presents an improved feature point screening algorithm with scale invariant feature transform(SIFT)and speeded up robust feature(SURF)based on the similar triangle principle for dune images to improve single sand dune tracking.The algorithm first uses a brutal force method to match the SIFT and SURF feature points,then filters the matching points according to the K nearest neighbor(KNN)algorithm,and finally uses the similar triangle principle to further filter the matching points.This method is then used to register the dunes in the Kumtag desert.The tests show that this algorithm is accurate and can be widely used.

关 键 词：沙丘跟踪 遥感图像 尺度不变特征变换(SIFT) 快速鲁棒特征(SURF) 图像配准 相似三角形 

分 类 号：S717.193[农业科学—林学]