基于改进SURF的图像特征点匹配算法  

Image feature point matching algorithm based on improved SURF

作　　者：王震洲[1] 张森 宁超 王建超 WANG Zhenzhou;ZHANG Sen;NING Chao;WANG Jianchao(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang,Hebei 050018,China)

机构地区：[1]河北科技大学信息科学与工程学院,河北石家庄050018

出　　处：《河北工业科技》2024年第6期418-425,共8页Hebei Journal of Industrial Science and Technology

基　　金：河北省教育厅青年基金(QN2023185);河北省高等学校科学技术重点研究项目(ZD2020318)。

摘　　要：为了提升图像在三维重建过程中特征点匹配时的准确率,提出了一种基于改进SURF(speeded-up robust features)的图像特征点匹配算法。首先,将SURF的64维度描述符提升至128维度;其次,在SURF算法引入KD-Tree模块,利用BBF(best bin first)最近邻查询机制实现特征点匹配;最后,通过对同一数据集进行旋转、缩小,并利用传统算法和改进SURF算法分别对图像进行特征点匹配实验,验证改进SURF算法的有效性。结果表明:改进SURF算法的特征点匹配正确率达到了89.19%,相较于传统SURF算法提高了17.62个百分点;特征错误匹配数由85减少至31,显著降低了特征点的匹配误差;运行时间由1.956 s缩短至1.647 s,进一步提升了算法的运行速度。改进后的SURF算法具备正确率高、误差小、速度快的特点,可为三维重建特征匹配工作提供重要的参考。In order to improve the accuracy of feature point matching in 3D reconstruction of images,an image feature point matching algorithm based on improved SURF was proposed.Firstly,the 64-dimension descriptor of SURF was upgraded to 128 dimensions.Secondly,the KD-Tree module was introduced into the SURF algorithm,and the BBF(best bin first)nearest neighbor query mechanism was used to implement feature point matching.Finally,after rotating and scaling changes on the same data set,feature point matching experiments were performed on the images by using the traditional algorithm and the improved SURF algorithm,respectively,and then the effectiveness of the improved SURF(speeded-up robust features)algorithm was verified.The results show that the improved SURF algorithm achieves a feature matching accuracy of 89.19%,which is improved 17.62 percentage points,and the number of feature false matches decreases from 85 to 31,significantly reducing the matching error of feature points.The running time is shortened from 1.956 seconds to 1.647 seconds,further improving the running speed of the algorithm.The improved SURF algorithm has the characteristics of high accuracy,less errors and fast speed,which can provide reference for 3D reconstruction feature matching.

关 键 词：图像处理 特征点匹配 SURF算法 KD-TREE 三维重建 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]