Fast Scene Matching Method Based on Scale Invariant Feature Transform
-
Institute of Instrument Science and Photoelectric Engineering, Beihang University, Beijing 100191, China
-
摘要:
傳統(tǒng)基于特征的景象匹配方法存在冗余點多、匹配精度低等問題，難以同時滿足實時性及魯棒性要求，對此，論文提出一種基于尺度不變特征變換(SIFT)的快速景象匹配方法。在特征提取階段，采用高速分段特征檢測器(FAST)在多尺度檢測角點作為初始特征，經(jīng)過高斯差分(DOG)算子在尺度空間中進行特征的2次篩選，簡化原有遍歷式的特征搜索過程；在特征匹配階段，采用仿射模型模擬變換關(guān)系建立幾何約束條件，克服SIFT算法由于忽略幾何信息而產(chǎn)生的誤匹配。實驗表明：該方法在匹配精度和實時性方面均優(yōu)于SIFT算法，且對光照、模糊、尺度等變換具有良好的魯棒性，能夠更好地實現(xiàn)景象匹配。
Abstract:The traditional feature-based image matching method has many problems such as many redundant points and low matching accuracy, which can hardly meet the real-time and robustness requirements. In this regard, a fast scene matching method based on Scale Invariant Feature Transform (SIFT) is proposed. In the feature detection phase, FAST (Features from Accelerated Segment Test) is used to detect characteristics in multi-scale, after then, combining with Difference Of Gauss (DOG) operators to filter characteristics again. From this, the feature search process is simplified. In feature matching phase, the affine transformation model is used to simulate the transformation relation and establish the geometric constraint, to overcome the mismatching because of ignoring the geometric information. The experimental results show that the proposed method is superior to the SIFT in efficiency and precision, also has good robustness to light, blur and scale transformation, achieves scene matching better.
-
表 1 相關(guān)性實驗數(shù)據(jù)
圖像 DOG特征點數(shù) FAST特征點數(shù) DOG∩FAST 重復(fù)率(%) Img1 2261 3531 627 27.7 Img2 1840 2276 418 22.7 Img3 3195 2168 711 32.8 Img4 2020 1523 473 23.4 Img5 8143 8615 3013 37.0 Img6 7788 9176 3060 39.3 Img7 2812 1491 573 38.4 Img8 3214 2003 645 32.2 
下載: 導(dǎo)出CSV
表 2 本文算法與SIFT算法消耗時間對比(ms)
數(shù)據(jù)集 SIFT SURF I-SIFT 特征檢測時間 特征匹配時間 總時間 特征檢測時間 特征匹配時間 總時間 特征檢測時間 特征匹配時間 總時間 graffiti 30574 4082 34656 7727 1021 8748 13866 2984 16850 bikes 16652 2779 19431 4098 884 4982 5216 1052 6268 boat 41933 6093 48026 15010 3235 18245 34397 7923 42320 leuven 14502 2248 16750 3458 487 3945 8518 1369 9887 average 25915 3801 29716 7573 1407 8980 15499 3332 18831
下載: 導(dǎo)出CSV
-
張聞宇, 李智, 王勇軍, 等. 基于CenSurE-star特征的無人機景象匹配算法[J]. 儀器儀表學(xué)報, 2017, 38(2): 462–470. doi: 10.19650/j.cnki.cjai.2017.02.02ZHANG Wenyu, LI Zhi, WANG Yongjun, et al. UAV scene matching algorithm based on CenSurE-star feature[J]. Chinese Journal of Scientific Instrument, 2017, 38(2): 462–470. doi: 10.19650/j.cnki.cjai.2017.02.02 BROWN L G. A survey of image registration techniques[J]. ACM Computing Surveys, 1992, 24(4): 325–376. doi: 10.1145/146370.146374 BABRI U M, TANVIR M, KHURSHID K, et al. Feature based correspondence: A comparative study on image matching algorithms[J]. International Journal of Advanced Computer Science & Applications, 2016, 7(3): 235–246. doi: 10.14569/IJACSA.2016.070329 SMITH S M and BRADY J M. SUSAN—A new approach to low level image processing[J]. International Journal of Computer Vision, 1997, 23(1): 45–78. doi: 10.1023/A:1007963824710 LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91–110. doi: 10.1023/B:VISI.0000029664.99615.94 李鶴宇, 王青. 一種具有實時性的SIFT特征提取算法[J]. 宇航學(xué)報, 2017, 38(8): 865–871. doi: 10.3873/j.issn.1000-1328.2017.08.011LI Heyu and WANG Qing. A real-time SIFT feature extraction algorithm[J]. Journal of Astronautics, 2017, 38(8): 865–871. doi: 10.3873/j.issn.1000-1328.2017.08.011 BAY H, ESS A, TUYTELAARS T, et al. Speeded-up robust features[J]. Computer Vision & Image Understanding, 2008, 110(3): 404–417. doi: 10.1016/j.cviu.2007.09.014 MATAS J, CHUM O, URBAN M, et al. Robust wide-baseline stereo from maximally stable extremal regions[J]. Image & Vision Computing, 2004, 22(10): 761–767. doi: 10.1016/j.imavis.2004.02.006 ROSTEN E and DRUMMOND T. Machine learning for high-speed corner detection[C] Springer-Verlag European Conference on Computer Vision, Graz, Austria, 2006: 430–443. 劉妍, 余淮, 楊文, 等. 利用SAR-FAST角點檢測的合成孔徑雷達圖像配準方法[J]. 電子與信息學(xué)報, 2017, 39(2): 430–436. doi: 10.11999/JEIT160386LIU Yan, YU Huai, YANG Wen, et al. SAR image registration sing SAR-FAST corner detection[J]. Journal of Electronics &Information Technology, 2017, 39(2): 430–436. doi: 10.11999/JEIT160386 常旭劍, 韓燮, 熊風(fēng)光, 等. 基于FAST檢測及SIFT描述的特征檢測算法[J]. 計算機工程與設(shè)計, 2015, 36(10): 2749–2753. doi: 10.16208/j.issn1000-7024.2015.10.028CHANG Xujian, HAN Xie, XIONG Fengguang, et al. Feature detection based on FAST detection and SIFT description[J]. Computer Engineering &Design, 2015, 36(10): 2749–2753. doi: 10.16208/j.issn1000-7024.2015.10.028 TONY L. Scale-space theory: A basic tool for analyzing structures at different scales[J]. Journal of Applied Statistics, 1994, 21(1/2): 225–270. doi: 10.1080/757582976 MIKOLAJCZYK K and SCHMID C. An affine invariant interest point detector[C]. Springer-Verlag European Conference on Computer Vision, Copenhagen, Denmark, 2002: 128–142. ELMASHAD S Y and SHOUKRY A. A more robust feature correspondence for more accurate image recognition[C]. IEEE Computer and Robot Vision. Montreal, Canada, 2014: 181–188. KANNALA J, RAHTU E, HEIKKILA J, et al. A new method for affine registration of images and point sets[J]. Lecture Notes in Computer Science, 2005, 40(35): 224–234. doi: 10.1007/11499145_25 王華夏, 程詠梅, 劉楠. 面向山地區(qū)域光照變化下的魯棒景象匹配方法[J]. 航空學(xué)報, 2017, 38(10): 188–200. doi: 10.7527/S1000-6893.2017.321101WANG Huaxia, CHENG Yongmei, LIU Nan. A robust scene matching method for mountainous regions with illumination variation[J]. Acta Aeronautica et Astronautica Sinica, 2017, 38(10): 188–200. doi: 10.7527/S1000-6893.2017.321101 -
計量
- 文章訪問數(shù): 1783
- HTML全文瀏覽量: 672
- PDF下載量: 76
- 被引次數(shù): 0
下載:
