基于SIFT的图像匹配算法  

Study of Image Matching Algorithm Based on SIFT

作　　者：贾世杰[1,2] 王鹏翔[1] 姜海洋[1] 曾洁[1] 

机构地区：[1]大连交通大学电气信息学院,辽宁大连116028 [2]大连理工大学电子信息与电气工程学部,辽宁大连116023

出　　处：《大连交通大学学报》2010年第4期17-21,共5页Journal of Dalian Jiaotong University

基　　金：国家中小企业创新基金资助项目(09C26222123243)

摘　　要：基于SIFT的图像匹配算法通过尺度空间检测特征点,计算特征点邻域的梯度方向直方图生成SIFT特征向量,然后采用SIFT特征向量的欧式距离来作为两幅图像中关键点的相似性度量.从不同阈值设置、视角变化及遮挡、添加噪声等几个方面对基于SIFT的图像匹配进行了研究和探讨.实验结果证明,本算法对图像尺度、视角变化、目标遮挡、噪声影响等方面有较好的鲁棒性,适用于在海量特征数据库中进行快速、准确的匹配.An image matching algorithm based on SIFT is implemented through detecting feature points in scale space,the histogram of the gradient direction of the feature points neighborhood is calculated to generate SIFT feature vector,using Euclidean distance between SIFT feature vectors as similarity measure of the two key points of the image.Several aspects of the SIFT-based image matching are studied and discussed,such as different threshold settings,perspectives change and clutterr,adding noise,and so on.Experimental results show that the algorithm has good robustness to the variance of image scale and perspective,clutter,noise,with fast and accurate matching of mass characteristics database.

关 键 词：图像匹配 SIFT 特征检测 特征描述 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]