计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (9): 141-145.DOI: 10.3778/j.issn.1002-8331.1510-0327

• 模式识别与人工智能 • 上一篇    下一篇

序列图像特征提取与匹配算法的改进

林  汀1,娄小平1,2,刘  锋1,2,李伟仙1,2   

  1. 1.北京信息科技大学 光电测试技术北京市重点实验室,北京 100192
    2.北京信息科技大学 光电信息与仪器北京市工程研究中心,北京 100192
  • 出版日期:2017-05-01 发布日期:2017-05-15

Improved  feature points  matching  algorithm on sequence image

LIN Ting1, LOU Xiaoping1,2, LIU Feng1,2, LI Weixian1,2   

  1. 1.Key Laboratory of Optoelectronic Measurement Technology, Beijing Information Science & Technology University, Beijing 100192, China
    2.Engineering Research Center for Optoelectronic Information and Instrument, Beijing Information Science & Technology University, Beijing 100192, China
  • Online:2017-05-01 Published:2017-05-15

摘要: 在计算机视觉领域中特征点匹配是一个重要课题。针对ORB(ORiented Brief,方向描述符)算法缺少尺度不变性的特点,将SURF(Speeded-Up?Robust?Features,快速鲁棒特征)算法与ORB相结合,提出了基于算法组合的改进算法SUORB(Speeded-Up?ORiented Brief,快速方向描述符)。组合算法的基本思路是利用SURF算法建立多尺度空间,然后通过ORB算法为检测出的特征点建立描述符,最后根据生成的二进制描述符实现特征点匹配。实验结果表明,SUORB算法基本弥补了ORB算法的不足,若图像尺度发生变化,SUORB匹配算法比ORB匹配算法的准确度明显提高;同时SUORB算法保留了ORB算法的快速性。

关键词: 特征点匹配, 多尺度空间, 组合算法, 方向描述符(ORB), 快速鲁棒特征(SURF)

Abstract: Feature point matching is an important subject in computer vision field. For the lack of scale invariance characteristics of ORB(ORiented Brief)algorithm, this paper proposes an improved method based on the combination of SURF(Speeded-Up?Robust?Features)and ORB, whose name is SUORB(Speeded-Up?ORiented Brief). Firstly, the scale spaces are built, in which the stable extreme points are detected in order to get the feature points with scale invariance information. Then, the feature points are described by the ORB descriptors with rotation invariance. Finally, Hamming distance is used to finish the final matching task. The experimental results show that SUORB has solved the deficiencies that ORB has little scale invariance. SUORB improves the matching accuracy, compared to ORB when images have scale changes. At the same time, the matching speeds of SUORB and ORB are almost the same.

Key words: feature points matching, multi scale space, combination of algorithms, ORiented Brief(ORB), Speeded-Up?Robust?Features(SURF)