中国空间科学技术

中国空间科学技术 ›› 2024, Vol. 44 ›› Issue (3): 157-166.doi: 10.16708/j.cnki.1000-758X.2024.0049

• 论文 • 上一篇    下一篇

基于非线性各向异性滤波的图像特征匹配算法

李华，杨杨，陈雨杰   

  1. 1.长春理工大学 计算机科学技术学院，长春130000
    2.特种电影技术及装备国家地方联合工程研究中心，长春130000
    3.数字媒体与虚拟现实实验室（长春理工大学），长春130000
  • 出版日期:2024-06-25 发布日期:2024-06-05

Image feature matching algorithm based on nonlinear anisotropic filtering

LI Hua，YANG Yang，CHEN Yujie   

  1. 1.College of Computer Science and Technology，Changchun University of Science and Technology，Changchun 130000，China
    2.National and Local Combined Engineering Research Center of Special Film Technology and Equipment，Changchun 130000，China
    3.Digital Media and Virtual Reality Laboratory（Changchun University of Science and Technology），Changchun 130000，China
  • Published:2024-06-25 Online:2024-06-05

PDF (PC)

44

摘要： 图像特征匹配是增强现实系统中的关键技术，匹配精度是提升特征匹配性能的关键。提出了一种多尺度特征匹配加强算法（I-AKAZE），通过对非线性各向异性滤波过程中传导函数的改进，减缓图像梯度值大的区域非线性扩散速度，极大程度地保留了匹配图像的边缘特征；同时，结合改进的非线性量化加速稳健特征描述符（NLG-SURF），提高了描述符的识别率。实验结果表明I-AKAZE算法在Mikolajczyk数据集上的可重复性得分相比目前先进的AKAZE算法有着大幅度提升，对应的特征描述符的平均识别率提升8.4%，并且运行速度比经典的SIFT算法快约19%，算法整体在检测和描述阶段上的性能都有提升。

关键词: 特征检测, 特征描述符, 非线性各向异性滤波, 尺度空间, 传导函数

Abstract:  Image matching is the key technology in augmented reality system，and matching accuracy is the key to improving the performance of feature matching.A multi-scale feature matching enhancement algorithm（I-AKAZE） is proposed.By improving the conduction function in the process of nonlinear anisotropic filtering，the nonlinear diffusion speed in the region with large gradient value of the image is slowed down，and the edge features of the matched image are retained to a great extent.At the same time，combined with the improved nonlinear quantization accelerated robust feature descriptor（NLG-SURF），the recognition rate of the descriptor is improved.The experimental results show that the repeatability score of I-AKAZE algorithm on Mikolajczyk data set is greatly improved compared with the current advanced AKAZE algorithm，that the average recognition rate of the corresponding feature descriptors is increased by 8.4%，and that the running speed is about 600ms faster than that of the classic SIFT algorithm.The overall performance of the algorithm is improved in the detection and description stages.

Key words:  , feature detection；feature descriptor；nonlinear filtering；scale space；conduction function