针对SIFT及其变体的图像匹配性能比较  

Performance Comparison of SIFT and its Variants for Image Matching

作　　者：栾广宇[1] 黄操军[1] 陈争光[1] 曹洪军[1] Luan Guangyu Huang Caojun Chen Zhengguang Cao Hongjun(Heilongjiang Bayi Agricultural University, Daqing 163319)

机构地区：[1]黑龙江八一农垦大学信息技术学院,大庆163319

出　　处：《黑龙江八一农垦大学学报》2017年第5期106-110,共5页journal of heilongjiang bayi agricultural university

基　　金：黑龙江省教育厅科学技术研究项目(12541584)

摘　　要：由于尺度不变特征变换算法(SIFT)及其变体(PCA-SIFT、GCSIFT、SURF、ASIFT)在图像匹配应用的广泛性,总结各自特点及比较它们性能具有实际的意义。首先系统地分析了SIFT及其变体,然后在尺度和旋转变化、光照变化、模糊变化、仿射变化四种图像情况评估了它们的性能。实验结果表明在尺度和旋转变化上SIFT执行最好;在光照和模糊变化上GCSIFT执行最好;在仿射变化上ASIFT表现最好;在所有情况下PCA-SIFT性能总是第二和SURF性能最差。Due to strong ability of SIFF and its variants （PCA-SIFT,GCSIFT,SURF and ASIFT） for image matching,that summarizing their characteristics and comparing their performance is meaningful. This paper first systematically analyzes SIFlr and its variants. Then,it evaluates their performance of the five algorithms in different situations that are scale and rotation change, illumination change, blur change and affine change. The experimental results show SIFT performs the best under scale and rotation change, GCSIFT performs the best under illumination change and blur change, ASIFT performs the best under affine change, PCA- SIFT is always the second and SURF performs the worst in different situations.

关 键 词：特征匹配 SIFT PCA-SIFT GCSIFT SURF ASIFT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]