一种改进的SIFT特征提取新算法  

An improved new algorithm for SIFT feature extraction

作　　者：苏丽颖[1] 李小鹏[1] 么立双[1] 

机构地区：[1]北京工业大学机械工程与应用电子技术学院,北京100124

出　　处：《华中科技大学学报（自然科学版）》2013年第S1期396-398,共3页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：教育部留学回国人员科研启动基金资助项目

摘　　要：针对尺度不变特征变换(SIFT)角点提取算法比较复杂、运算速度比较慢的缺点,提出了一种更加快速的SIFT角点提取方法.通过对SIFT算法的分析,提出了一种候选角点筛选与SIFT算法相结合的新算法.首先,应用像素点8邻域相似度方法对要提取角点的图像进行候选角点筛选,然后以候选角点为中心构建3×3的灰度图像,再应用SIFT提取算法对构建的3×3的灰度图像进行角点提取.通过实验和理论推导证明了新算法的有效性.A more rapid SIFT(scale invariant feature transform)corner extraction method was proposed to overcome the shortcomings of the existing complex SIFT corner extraction algorithm and relative slow running speed.A new algorithm named combination of candidate screening and SIFT algorithm was proposed based on the SIFT algorithm analysis.Firstly,corner candidate screening was carried out for the image of desired extracting corners by applying 8neighborhood similarity method, and 3×3gray image was built centering on the candidate corner.Then,SIFT extraction algorithm was applied to extract candidate corners for the built 3×3gray image.The validity of the new algorithm was verified through the experiment and theoretical derivation.

关 键 词：SIFT算法 候选角点筛选 像素点 灰度图像 8邻域 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]