基于改进ORB的图像特征点匹配研究  

Research of image feature point matching based on improved ORB algorithm

作　　者：戴雪梅[1] 郎朗[1] 陈孟元[1] 

机构地区：[1]安徽工程大学安徽省电气传动与控制重点实验室,芜湖241000

出　　处：《电子测量与仪器学报》2016年第2期233-240,共8页Journal of Electronic Measurement and Instrumentation

基　　金：2015年度安徽省电气传动与控制重点实验室开放基金课题(201503)项目资助

摘　　要：针对ORB算法不具备尺度不变性的缺陷,结合SURF算法,将ORB算法改进为SURB(SURF-ORB)算法。首先提取具有尺度不变性的SURF特征点,然后构建ORB算法描述子,特征匹配时,先对待处理图像进行区域分块,缩减原始搜索范围,节约了无效区域的特征点匹配时间,采用汉明距离完成初步匹配,然后结合PROSAC算法对初步筛选的点去伪,获得较为精准的匹配点对。实验结果表明,图像尺度变化时,SURB算法的平均匹配准确度为93.4%,约为ORB算法的3倍;SURB算法的平均耗时约为SURF算法的80%,具有较强的实时性和可行性。In view of ORB algorithm does not have scale invariant,the ORB algorithm is improved combined with SURF algorithm,SURB（ SURF-ORB） algorithm is formed. Firstly,the feature points of SURF are extracted and then ORB descriptor is built. During feature matching,the matching image is divided into two parts,and the image search range is reduced. Then,Hamming distance completes the coarse matching,which avoids global feature points extraction and matching. Finally,false matching points are removed by PROSAC algorithm to improve matching accuracy. The experimental results show that the average matching accuracy of SURB algorithm is 93. 4%,which is approximately 3 times that of ORB algorithm. The average time-consuming of SURB is about 80% of SURF algorithm,which can meet the needs of real-time application.

关 键 词：特征点匹配 尺度不变性 ORB SURB 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN957.52[自动化与计算机技术—计算机科学与技术]