计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (19): 208-212.

• 图形图像处理 • 上一篇    下一篇

基于快速视网膜关键点算法改进的图像匹配方法

付  偲1,2,邓  丽1,2,卢  根1,2,费敏锐1,2   

  1. 1.上海大学 机电工程与自动化学院,上海 200072
    2.上海市电站自动化技术重点实验室,上海 200072
  • 出版日期:2016-10-01 发布日期:2016-11-18

Improved image matching based on fast retina keypoint algorithm

FU Cai1,2, DENG Li1,2, LU Gen1,2, FEI Minrui1,2   

  1. 1.School of Mechatronical Engineering and Automation, Shanghai University, Shanghai 200072, China
    2.Shanghai Key Laboratory of Power Station Automation Technology, Shanghai 200072, China
  • Online:2016-10-01 Published:2016-11-18

摘要: 传统的仿射尺度不变特征(ASIFT)算法通过模拟仿射变化图像实现完全仿射不变性,但是由于尺度不变特征(SIFT)算法本身的低效造成ASIFT的过程非常耗时,为了实现更为高效的图像匹配,引入快速视网膜关键点(FREAK)算法到ASIFT仿射模型中,并基于Lanczos-4插值进行改进。在匹配过程中基于HAMMING距离实现暴力匹配,并结合随机样本一致性(RANSAC)算法改进对匹配点对的提纯,得到了新的AFREAK算法。该算法既能实现完全仿射不变性,又能实现低耗时和低内存占用。实验结果表明,提出的AFREAK算法处理速度上快于ASIFT近2~3倍,并且可以得到与之相似的匹配效果。

关键词: 尺度不变特征(SIFT), 仿射尺度不变特征(ASIFT), 快速视网膜关键点算法(FREAK), 仿射不变, 图像匹配

Abstract: Conventional Affine Scale Invariant Features(ASIFT) algorithm implements full affine invariance by simulating image with affine transformation. To solve the problem of time-consuming implements caused by low efficiency of SIFT algorithm and implement more efficient image matching, Fast Retina Keypoint(FREAK) is introduced to the affine model of ASIFT with improvements based on Lanczos-4 interpolation. With the implementation of Brute Force feature matching based on HAMMING distance and improvement of matching points pairs filtration combined with Random Sample Consensus(RANSAC), the new algorithm AFREAK is obtained, which implements full affine invariance with low consuming and memory usage. Experimental results show that the speed of proposed algorithm is almost 2 to 3 times faster than the original ASIFT algorithm with the similar matching effect.

Key words: Scale Invariant Features(SIFT), Affine Scale Invariant Features(ASIFT), Fast Retina Keypoint(FREAK), affine invariance, image matching