基于改进FREAK的图像匹配算法  

Image Matching Algorithm Based on Improved FREAK

作　　者：黄坤 钱军浩[1] 王江文[2] HUANG Kun;QIAN Junhao;WANG Jiangwen(School of Internet of Things Engineering,Jiangnan University,Wuxi,Jiangsu 214122,China;State Key Laboratory of Traction Power,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122 [2]西南交通大学牵引动力国家重点实验室,成都610031

出　　处：《计算机工程》2020年第8期250-257,共8页Computer Engineering

基　　金：国家自然科学基金(51475391)。

摘　　要：为提升图像匹配算法的实时性和鲁棒性,提出一种基于改进FREAK的特征点匹配算法。将经典FREAK算法的8层视网膜模型简化为5层,根据贪婪搜索算法选取64组感受野点对,以在减少运算开销的同时尽量保留有效的点对信息。在此基础上,设计一种具有方向不变性的LBP算法对每个感受野进行编码,从而提高描述符的区分度。实验结果表明,与FREAK、BRISK等算法相比,该算法具有最小的描述符尺寸,且在多数场景下,其运算更快,精度更高,更适合光照变化复杂的环境。To improve the real-time performance and robustness of image matching algorithms,this paper proposes a feature point matching algorithm based on improved FREAK.The algorithm simplifies the 8-layer retina model of the classical FREAK algorithm to a 5-layer one,and uses a greedy search algorithm to select 64 groups of receptive field pairs,so as to reduce the overhead of calculation and keep as much useful point pair information as possible.On this basis,a rotation-invariant Local Binary Patterns(LBP)algorithm is designed to encode every receptive field in order to increase the discriminative power of the descriptor.Experimental results show that compared with FREAK,BRISK and other algorithms,the proposed algorithm has the smallest descriptor size,and in most scenes has a higher calculation speed and accuracy,which means it is more suitable for environments with complex light changes.

关 键 词：图像匹配 FREAK算法 局部二值模式 特征点匹配 描述符 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]