一种CSS-SIFT复合图像配准算法  

CSS-SIFT Composite Image Registration Algorithm

作　　者：李培华 章盛 刘玉莉 钱名思 LI Peihua;ZHANG Sheng;LIU Yuli;QIAN Mingsi(AVIC Huadong Photoelectric Co.,Ltd.,Wuhu 241002,China;Key Laboratory of Modern Display Technology,Wuhu 241002,China;National Special Display Engineering Research Center,Wuhu 241002,China;National Engineering Laboratory of Special Display Technology,Wuhu 241002,China)

机构地区：[1]中航华东光电有限公司,安徽芜湖241002 [2]安徽省现代显示技术重点实验室,安徽芜湖241002 [3]国家特种显示工程技术研究中心,安徽芜湖241002 [4]特种显示国家工程实验室,安徽芜湖241002

出　　处：《红外技术》2021年第1期26-36,共11页Infrared Technology

基　　金：安徽省科技重大专项项目(17030901053)。

摘　　要：针对SIFT算法的图像配准耗时长的问题,提出一种CSS-SIFT复合图像配准算法。CSS-SIFT算法首先使用CSS算法检测图像特征,然后,使用优化的SIFT算法生成并降维图像特征描述子,最后,使用基于欧式距离和曼哈顿距离的优化双向匹配算法对图像特征进行匹配。仿真实验条件是通过计算机中仿真软件进行仿真实验,统计图像特征数目、匹配数目、正确匹配数目、配准准确率、配准时间与配准时间下降率共6个指标数据,统计结果表明,CSS-SIFT算法在图像配准准确度方面与传统SIFT算法、传统SURF算法、Forstern-SIFT算法、Harris-SIFT算法、Trajkovic-SIFT算法相当,但在图像配准耗时方面分别降低了58.45%、10.68%、14.84%、16.21%与4.63%,为图像配准提供了一种解决方案。To address the time-consuming problem of image registration in the scale-invariant feature transform(SIFT)algorithm,a curvature scale space(CSS)-SIFT composite image registration algorithm is proposed in this paper.First,the CSS-SIFT algorithm uses the CSS algorithm to extract image features.Image feature descriptors are then generated and reduced by the optimized SIFT algorithm.Finally,an optimized two-way matching algorithm based on Euclidean and Manhattan distances is used for matching.A simulation experiment is conducted using simulation software,and six parameters of index data are employed,including the number of image features,number of matches,number of correct matches,registration accuracy,registration time,and registration time decline rate.Statistical results show that the CSS-SIFT algorithm performs as well as the following algorithms in terms of accuracy of image registration:traditional SIFT,traditional speeded-up robust features,Forstern-SIFT,Harris-SIFT,and Trajkovic-SIFT.In addition,time-consumption of image registration is reduced by 58.45%,10.68%,14.84%,16.21%,and 4.63%,respectively,thus providing an effective solution for image registration.

关 键 词：尺度不变特征变换算法 加速稳健特征算法 曲率尺度空间算法 图像配准 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]