高分辨率遥感图像SIFT和SURF算法匹配性能研究

齐冰洁; 刘金国; 张博研; 左洋; 吕世良

doi:10.3788/CO.20171003.0331

高分辨率遥感图像SIFT和SURF算法匹配性能研究

doi: 10.3788/CO.20171003.0331

1.
中国科学院长春光学精密机械与物理研究所, 吉林长春 130033
2.
中国科学院大学, 北京 100049

基金项目:

国家自然科学基金资助项目 61405191

详细信息

作者简介:
齐冰洁(1991-), 女, 辽宁开原人, 硕士研究生, 主要从事遥感图像处理方面的研究。E-mail:doudouqbj@126.com

刘金国(1968—)，男，吉林蛟河人，硕士，研究员，硕士生导师，主要从事空间遥感与图像处理方面的研究。E-mail: liujg@ciomp.ac.cn

通讯作者:
刘金国, E-mail: liujg@ciomp.ac.cn

中图分类号: TP391
计量
- 文章访问数: 1974
- HTML全文浏览量: 491
- PDF下载量: 700
- 被引次数: 0
出版历程
- 收稿日期: 2017-01-18
- 修回日期: 2017-03-27
- 刊出日期: 2017-06-01

Research on matching performance of SIFT and SURF algorithms for high resolution remote sensing image

1.
Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China

Funds:

National Natural Science Foundation of China 61405191

摘要

摘要: 遥感图像匹配是图像校正、拼接的基础。由于遥感图像特征相似度大，重叠区域小，遥感图像对匹配算法的要求更高。本文首先从特征检测、特征描述和特征匹配三个方面，比较了SIFT算法和SURF算法在计算速度和准确度方面性能，然后研究了算法对遥感图像重叠度、度量距离的要求，并针对SURF算法对特征方向误差敏感的特点，提出一种oSURF算法；最后利用卫星1A级条带遥感图像分析各个算法优劣性。测试结果表明，相比于SIFT算法，SURF算法计算速度为SIFT的3倍，需要的图像重叠宽度仅为1.25倍描述向量尺寸，而在保证同样匹配率的情况下，SIFT算法则需要图像重叠宽度为1.5倍描述向量尺寸。本文提出的oSURF算法在保证计算速度的同时，准确度相对于SURF算法提升5%~10%，因此，oSURF算法更适合1A级条带遥感图像的拼接。
- 遥感图像 /
- 特征匹配 /
- SIFT /
- SURF /
- 重叠度
Abstract: Image matching is the basis of image rectification and mosaic. Because of higher features similarity and smaller overlapped area than ordinary images, the remote sensing images have higher requirements on matching algorithm in both performance and iteration speed. The performances in three aspects:feature detection, feature description and feature matching, are analyzed between the SIFT algorithm and the SURF algorithm in terms of speed and accuracy. The requirements of the degree of overlapping between remote sensing images and the matching distance of the genvector is discussed as well. In view of the characteristic that SURF algorithm is sensitive to the error in feature detection, oSURF algorithm is presented in this paper. Finally, the advantages and disadvantages of each algorithm are analyzed by using satellite remote sensing data of level 1A. The results show that iteration speed of SURF algorithm is three times faster than SIFT algorithm. Under the same matching rate, the width of overlapped area on image required in SURF algorithm is 1.25 times of the dimension of genvector but 1.5 times in SIFT, and the accuracy of oSURF algorithm is increased by 5%~10% compared with SURF algorithm in the same computation speed, which indicate that oSURF is more suitable for remote sensing image stitching.
- remote sensing image /
- feature matching /
- SIFT /
- SURF /
- degree of overlapping

HTML全文

图 1 美国某海岸影像数据

Figure 1. Image of an America′s shore

下载: 全尺寸图片幻灯片

图 2 SIFT和SURF算法特征点检测效果

Figure 2. Detection result of SIFT and SURF algorithm

下载: 全尺寸图片幻灯片

图 3 检测区域宽度不同时描述向量个数变化

Figure 3. Vector number change with the detection area

下载: 全尺寸图片幻灯片

图 4 重叠区域不同时同名点概率变化

Figure 4. Probability change of Homologous Points with the overlap region

下载: 全尺寸图片幻灯片

图 5 不同算法匹配效果图

Figure 5. Matching result of different algorithms

下载: 全尺寸图片幻灯片

图 6 非重叠区域的影响

Figure 6. Influence of non-overlapping domain

下载: 全尺寸图片幻灯片

图 7 不同算法计算距离不同时正确率的变化情况

Figure 7. Matching ratio change with the calculated distance

下载: 全尺寸图片幻灯片

图 8 Haar小波滤波器

Figure 8. Haar wavelet filters

下载: 全尺寸图片幻灯片

图 9 扇形窗口

Figure 9. Sector window

下载: 全尺寸图片幻灯片

图 10 oSURF算法匹配效果图

Figure 10. Matching result of oSURF algorithm

下载: 全尺寸图片幻灯片

图 11 不同算法匹配结果

Figure 11. Matching result of different algorithms

下载: 全尺寸图片幻灯片

表 1 SIFT和SURF算法特征检测结果

Table 1. Detection result of SIFT and SURF algorithm

图像	算法	特征点数/个	检测时间/s
条带图像1	SIFT	438	5.013
	SURF	523	1.596
条带图像2	SIFT	480	5.326
	SURF	573	1.869

下载: 导出CSV

表 2 不同算法特征向量描述结果

Table 2. Feature description result of different algorithms

算法	图像	特征向量	特征点变化	描述时间	10个向量描述时间
SIFT	条带图像1	1 119	+681	12.962	0.116
	条带图像2	1 451	+971	13.863	0.096
SURF64	条带图像1	464	-59	3.015	0.065
	条带图像2	520	-53	3.368	0.065
SURF128	条带图像1	464	-59	3.213	0.069
	条带图像2	520	-53	3.473	0.067

下载: 导出CSV

表 3 不同算法特征匹配结果比较

Table 3. Matching result of different algorithms

算法	匹配点	正确率	匹配时间	总时间
SIFT	286	0.979	1.152	38.521
SURF64	84	0.896	0.310	11.514
SURF128	120	0.946 7	0.425	12.581

下载: 导出CSV

表 4 oSURF算法特征匹配结果比较

Table 4. Matching result of oSURF algorithm

算法	条带1描述向量	条带2描述向量	匹配点数	准确度	总时间
oSURF	814	963	289	0.965 4	12.926

下载: 导出CSV

参考文献(20)

[1]	LOWE D G. Object recognition from local scale-invariant features[C]. International Conference on Computer Vision, Kerkyra, Greece:ICCV, 1999:1150-1157.
[2]	LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International J. Compute Vision, 2004, 60(60):91-110. http://nichol.as/papers/Lowe/Distinctive%20Image%20Features%20from%20Scale-Invariant.pdf
[3]	BAY H, TUYTELAARS T, GOOL L. SURF:speeded up robust features[J]. Computer Vision & Image Understanding, 2006, 110(3):404-417. http://dev.ipol.im/~morel/Dossier_MVA_2010_Cours_Transparents_Documents/Cours_4_document_SURF_paper.pdf
[4]	BAY H, ESS A, TUYTELAARS T, et al.. Speeded-Up Robust Features (SURF)[J]. Computer Vision & Image Understanding, 2008, 110(3):346-359. http://www.micc.unifi.it/delbimbo/wp-content/uploads/2011/03/reference_papers/Bay.pdf
[5]	LUO J, GWUN O. A Comparison of SIFT, PCA-SIFT and SURF[J]. J. Business Education, 2009, 3(4):143-152.
[6]	索春宝, 杨东清, 刘云鹏.多种角度比较SIFT、SURF、BRISK、ORB、FREAK算法[J].北京测绘, 2014, (4):23-26. http://www.cnki.com.cn/Article/CJFDTOTAL-BJCH201404006.htm SUO CH B, YANG D Q, LIU Y P. Comparison of SIFT, SURF, BRISK, ORB and FREAK algorithms in various aspects[J]. Beijing Surveying and Mapping, 2014, (4):23-26.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-BJCH201404006.htm
[7]	安维胜, 余让明, 伍玉铃.基于FAST和SURF的图像配准算法[J].计算机工程, 2015, 41(10):232-235. doi: 10.3969/j.issn.1000-3428.2015.10.043 AN W SH, YU R M, WU Y L. Image registration algorithm based on FAST and SURF[J]. Computer Engineering, 2015, 41(10):232-235.(in Chinese) doi: 10.3969/j.issn.1000-3428.2015.10.043
[8]	王新华, 黄玮, 欧阳继红.多探测器拼接成像系统实时图像配准[J].中国光学, 2015, 8(2):211-219. http://www.chineseoptics.net.cn/CN/abstract/abstract9257.shtml WANG X H, HUANG W, OUYANG J H. Real-time image registration of the multi-detectors mosaic imaging system[J]. Chinese Optics, 2015, 8(2):211-219.(in Chinese) http://www.chineseoptics.net.cn/CN/abstract/abstract9257.shtml
[9]	许佳佳.结合Harris与SIFT算子的图像快速配准算法[J].中国光学, 2015, 8(4):574-581. http://www.chineseoptics.net.cn/CN/abstract/abstract9285.shtml XU J J. Fast image registration method based on Harris and SIFT algorithm[J]. Chinese Optics, 2015, 8(4):574-581.(in Chinese) http://www.chineseoptics.net.cn/CN/abstract/abstract9285.shtml
[10]	张云生, 邹峥嵘.基于改进ORB算法的遥感图像自动配准方法[J].国土资源遥感, 2013, 25(3):20-24. doi: 10.6046/gtzyyg.2013.03.04 ZAHNG Y SH, ZOU ZH R. Automatic registration method for remote sensing images based on improved ORB algorithm[J]. Remote Sensing for Land and Resources, 2013, 25(3):20-24.(in Chinese) doi: 10.6046/gtzyyg.2013.03.04
[11]	罗楠, 孙权森, 耿蕾蕾, 等.一种扩展SURF描述符及其在遥感图像配准中的应用[J].测绘学报, 2013, 42(3):383-388. http://www.cnki.com.cn/Article/CJFDTOTAL-CHXB201303012.htm LUO N, SUN Q S, GENG L L, et al.. An extended SURF descriptor and its application in remote sensing images registration[J]. Acta Geodaetica et Cartographica Sinica, 2013, 42(3):383-388.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-CHXB201303012.htm
[12]	闫利, 陈林.一种改进的SURF及其在遥感影像匹配中的应用[J].武汉大学学报:信息科学版, 2013, 38(7):770-773. http://www.cnki.com.cn/Article/CJFDTOTAL-WHCH201307005.htm YAN L, CHEN L. A modified SURF descriptor and its application in remote sensing images matching[J]. Geomatics and Information Science of Wuhan University, 2013, 38(7):770-773.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-WHCH201307005.htm
[13]	葛盼盼, 陈强.基于SURF特征提取的遥感图像自动配准[J].计算机系统应用, 2014, 23(3):16-24. http://www.cnki.com.cn/Article/CJFDTOTAL-XTYY201403003.htm GE P P, CHEN Q. Remote sensing image automatic registration based on SURF feature extraction[J]. Computer Systems & Applications, 2014, 23(3):16-24.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-XTYY201403003.htm
[14]	LINDEBERG T. Scale-space theory:a basic tool for analyzing structures at different scales[J]. J. Applied Statistics, 1994, 21(1):225-270. doi: 10.1080/757582976
[15]	VIOLA P, JONES M. Rapid object detection using a boosted cascade of simple features[C]. Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Kauai, HI, USA, 2001:511-518
[16]	翟优, 曾峦, 熊伟.不同局部邻域划分加速鲁棒特征描述符的性能分析[J].光学精密工程, 2013, 21(9):2395-2404. http://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201309026.htm ZHAI Y, ZENG L, XIONG W. Performance analysis of SURF descriptor with different local region partitions[J]. Opt. Precision Eng., 2013, 21(9):2395-2404.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201309026.htm
[17]	孙浩, 王程, 王润生.局部不变特征综述[J].中国图象图形学报, 2011, 16(2):141-151. doi: 10.11834/jig.20110207 SUN H, WANG CH, WANG R SH. A review of local invariant features[J]. J. Image and Graphics, 2011, 16(2):141-151.(in Chinese) doi: 10.11834/jig.20110207
[18]	聂海涛, 龙科慧, 马军, 等.采用改进尺度不变特征变换在多变背景下实现快速目标识别[J].光学精密工程, 2015, 23(8):2349-2356. http://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201508029.htm NIE H T, LONG K H, MA J, et al.. Fast object recognition under multiple varying background using improved SIFT method[J].Opt. Precision Eng., 2015, 23(8):2349-2356.(in Chinese) http://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201508029.htm
[19]	MIKOLAJCZYK K, SCHMID C. A performance evaluation of local descriptors[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10):1615-1630. doi: 10.1109/TPAMI.2005.188
[20]	FISCHLER M, BOLLES R. Random Sample Consensus:A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography[M]. Burlington:Morgan Kaufmann Publishers Inc., 1987.