改进的SURF-RANSAC图像匹配算法  

Improved SURF-RANSAC image matching algorithm

作　　者：赵谦[1] 童申鑫 贺顺[1] 尹怡晨 ZHAO Qian;TONG Shen-xin;HE Shun;YIN Yi-chen(School of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710054,China;Xi’an Dishan Vision Technology Limited Company,Xi’an 712044,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710054 [2]西安地山视聚科技有限公司,陕西西安712044

出　　处：《计算机工程与设计》2021年第10期2902-2909,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61801373);陕西省科技计划工业科技攻关基金项目(2017GY-073);西安市科技计划高校人才服务企业基金项目(2019217714GXRC013CG014-GXYD13.5)。

摘　　要：为改善传统SURF图像匹配算法存在计算量大、配准精度低等不足,提出一种改进SURF-RANSAC算法。基于SURF算法中构建描述符的思想,采用圆形邻域代替矩形邻域提取32维描述符,实现描述符的降维;通过自适应阈值方法完成特征点初匹配,降低人为设定阈值对匹配结果的影响;通过特征向量构建余弦约束对随机采样一致性(RANSAC)算法进行改进,实现匹配点对的提纯。实验结果表明,改进后的算法不仅提高了匹配速度和精度,还具有较强的鲁棒性。An improved SURF-RANSAC algorithm was proposed to improve the traditional SURF matching algorithm,which has the problems of large amount of calculation and low registration accuracy.Based on the idea of constructing descriptors in the SURF algorithm,a circular neighborhood was used instead of a rectangular neighborhood to extract 32-dimensional descriptors to achieve dimensionality reduction of the descriptors.The initial matching of feature points was completed using the adaptive threshold method,which reduced the impact of artificially set thresholds on the matching results.The random sampling consistency(RANSAC)algorithm was improved by constructing the cosine constraint of the feature vector to realize the purification of matching point pairs.Experimental results show that the improved algorithm improves not only the matching speed and accuracy,but its robustness.

关 键 词：SURF算法 降维 自适应阈值 余弦约束 随机采样一致性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]