一种具有强实时性、强鲁棒性的图像匹配算法  

A Strong Robust Real-Time Image Matching Algorithm

作　　者：李兵[1,2,3,4] 刘磊[5] 魏志强[6] 

机构地区：[1]新奥特(北京)视频技术有限公司,北京100195 [2]中国传媒大学计算机学院,北京100024 [3]中关村科技园区海淀园企业博士后,北京100089 [4]中国科学院自动化研究所,北京100190 [5]北京科技大学计算机与通信工程学院,北京100083 [6]北京航空航天大学机器人研究所,北京100191

出　　处：《软件学报》2014年第7期1583-1592,共10页Journal of Software

基　　金：中央高校基本科研业务费(FRF-TP-12-081A);北京市自然科学基金(4122049)

摘　　要：针对描述符BRIEF对图像旋转敏感的问题,提出一种改进的描述符RIBRIEF,该描述符具有识别能力强、提取速度快、占用空间小及抗干扰能力强等优点,并具有旋转不变性.经分析,图像匹配算法的实时性较大程度上由特征点数量、匹配点搜索次数及描述符相似度计算复杂度决定,因此提出通过描述符索引与描述符聚类相结合、基于FAST稳定特征点提取和逻辑运算计算相似度等方法提高算法的整体实时性.实验结果表明,与描述符BRIEF及SURF相比较,基于描述符RIBRIEF的图像匹配算法在鲁棒性及实时性方面均具有明显优势.To overcome the shortcomings of descriptor BRIEF which is sensitive to image rotation, this paper proposes a improved descriptor RIBRIEF which has the advantages of good identification ability, high descriptor extraction speed, less memory usage, strong robustness and rotation invariant. The study shows that real-time performance of image matching algorithm is largely decided by the number of feature points, the search times of matching points and the computational complexity of descriptor similarity. It therefore proposes optimization algorithms to improve real-time performance of image matching by combining descriptor index and descriptor cluster, applying FAST to stable feature point extraction and calculating descriptor similarity with logic operations. Compared with SURF and BRIEF, experimental results show that RIBRIEF has better performance in robustness and real-time.

关 键 词：RIBRIEF BRIFE FAST 索引 聚类 实时性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]