尺度不变特征转换特征提取优化算法研究  

Optimization of scale-invariant feature transform(SIFT) feature extraction

作　　者：易军凯[1] 边锆辉[1] 姜大光[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029

出　　处：《北京化工大学学报（自然科学版）》2013年第1期115-119,共5页Journal of Beijing University of Chemical Technology(Natural Science Edition)

基　　金：国家"863"计划重点项目(2009AA01Z433)

摘　　要：针对尺度不变特征转换(SIFT)算法时间复杂度高的缺点,提出了SIFT特征提取优化算法。分析了SIFT特征提取算法各个计算步骤的时间复杂性。对SIFT特征提取过程进行了优化,包括耗时最多的高斯金字塔的创建和计算特征描述符过程。优化算法降低了特征点提取时间,减少了特征点的重复匹配,同时保证了匹配结果的准确性。最后,实验证明了优化后的算法能有效降低时间复杂度。Given the limitations imposed by the high time complexity of scale-invariant feature transform （SIFT） al- gorithms, an optimized SIFT feature extraction algorithm has been developed. The time complexity of the calcula- tion steps of the SIFT feature extraction algorithm was first analyzed. Secondly, the Gaussian pyramid creation and feature^point calculation process, which are the most time-consuming processes in a SIFT algorithm, were opti- mized. The optimized algorithm succeeded in reducing the feature point extraction time and reducing the duplicated feature matching, whilst at the same time guaranteeing the accuracy of the matching results. Finally, experiments showed that the optimized algorithm effectively reduced the time complexity.

关 键 词：针对尺度不变特征转换算法 高斯金字塔 高斯核 特征描述符 特征点提取 物体识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]