基于边缘角点的SIFT图像配准算法  

Image Registration Algorithm of SIFT Based on Edge and Corner Point

作　　者：赵萌萌[1] 曹建秋[1] 

机构地区：[1]重庆交通大学信息科学与工程学院,重庆400074

出　　处：《重庆交通大学学报（自然科学版）》2013年第4期721-724,共4页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：重庆市科委攻关项目(CSTC 2011AC6102);交通运输部西部项目(20113188141480);重庆交通大学创新基金项目〔2010(下)第16号〕

摘　　要：SIFT由特征提取,特征描述符描述和特征匹配3部分构成,该算子特征提取数目庞大,建立特征描述符运算量高,导致算法效率低。提出了一种SEC(SIFT-Edge-Corner)算法,在图像尺度空间提取角点代替SIFT特征点,并根据角点是边缘曲率极值理论,预先采用Canny算子得到高斯边缘图像金字塔,再提取角点并进行尺度选择。实验结果表明:该算法在保障高准确率的前提下大幅度提高特征提取效率。SIFT is composed of feature extraction,feature descriptor representation and feature matching,which has a huge quantity in feature extraction and a high computational complexity in descriptor representation.So this algorithm is of low efficiency.In order to solve this problem,a kind of algorithm which is named SEC(SIFT-Edge-Corner) is proposed.Corner points instead of SIFT feature points are extracted in image scale space.According to the theory that corner is the extreme curvature of edge,SEC constructs edge-of-Gaussian pyramid by canny operator before extracting corner points,and then the scales are selected.The experimental results show that SEC improves the feature extraction efficiency at the premise of ensuring high accuracy.

关 键 词：SIFT 边缘提取 角点 SEC 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]