SIFT与SURF特征提取算法在图像匹配中的应用对比研究
2018-04-04陈敏汤晓安
陈敏 汤晓安
摘 要: SIFT算法和SURF算法是图像特征提取和匹配的典型方法,广泛应用于目标检测、图像理解与识别等领域,然而对二者尚缺乏较深入的对比研究。针对这两种算法,采用实验室相机实拍图像和低空无人机实拍图像,以不同的图像旋转角度进行特征点提取和图像匹配实验,从匹配成功率和运行效率两个方面对算法的性能进行对比研究。结果表明,SIFT算法具有较好的图像旋转不变性,匹配精度较高,而SURF算法匹配精度较低,但是效率较高,因此在实际应用中可根据具体需求合理选择。
关键词: SIFT; SURF; 性能对比; 特征提取; 图像匹配; 算法效率
中图分类号: TN911.73?34; TP391.6 文献标识码: A 文章编号: 1004?373X(2018)07?0041?04
Comparison study on application of SIFT and SURF feature
extraction algorithms in image matching
CHEN Min1, TANG Xiaoan2
(1. School of Electronic Information, Hunan Institute of Information Technology, Changsha 410151, China;
2. College of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073, China)
Abstract: The SIFT and SURF algorithms as two typical algorithms for image feature extraction and matching are widely used in the fields of target detection, image understanding and recognition, but the comparison research on the two algorithms is shallow. In allusion to the two algorithms, the feature point extraction and image matching experiment for the images taken by laboratory camera and UVA are carried out according to different image rotation angles. The performances of the two algorithms are compared in the aspects of matching success rate and operating efficiency. The results show that the SIFT algorithm has perfect image rotation invariance and high matching accuracy, and the SURF algorithm has low matching accuracy and high efficiency. One of the two algorithms can be selected reasonably according to the specific requirements.
Keywords: SIFT; SURF; performance comparison; feature extraction; image matching; algorithm efficiency
0 引 言
图像匹配是图像处理研究领域的重要问题,在计算机视觉、医疗图像分析、无人机图像处理等领域有着广泛应用。而图像特征提取是图像匹配的基础和前提,图像特征提取的效率、準确度等性能的好坏直接影响着后续的图像匹配[1?3]。例如,待匹配图像之间普遍存在缩放、旋转等现象,因此在特征提取中需要考虑特征提取的稳定性和准确性,而提取尺度不变的图像特征是解决这一问题的方法。
近年来,很多尺度不变特征的提取算法被提出和应用,具有代表性的是David G. Lowe提出的SIFT算法[4]以及Herbert Bay等人提出的SURF[5]算法。相关学者对这两种经典算法在不同的应用场合有了一定的对比研究[6?10],但是在无人机应用,特别是低空无人机影像的配准上尚缺少有实际意义的成果。本文给出这两种算法的原理,并基于两种不同的环境——实验室环境和无人机低空拍摄所获取的实拍图像进行特征点提取和图像匹配实验,结合实验结果比较两种算法的性能。……
