基于SIFT二次匹配方法的同名像点识别

摘要

The SIFT twice matching algorithm is proposed to get much more and reliable feature points in IRS-P5 remote sensing image matching. In order to get initial positioning points, uniqueness constraint and global geometric constraints are introduced to reduce matching error. Micro-region images are got according to initial positioning points, and a new constraint-offset coordinate constraint is proposed together with the uniqueness constraint to complete the SIFT twice matching. Experimental results show that the matching number using the SIFT twice matching algorithm is 23.07 times of using SIFT directly with strict threshold value.%将尺度不变特征变换(SIFT)二次匹配方法用于IRS-P5立体像对的同名像点识别.引入全局几何约束与唯一性约束,剔除误匹配,获取用于初始定位的匹配样本,完成初始匹配.根据初始定位点,获取小区域子图像,在小区域内调整SIFT匹配阈值,在唯一性约束基础上,引入偏移坐标差值约束,完成二次匹配.通过实验验证,相比于将SIFT算法直接应用于遥感影像同名像点识别,SIFT特征二次匹配算法在严格阈值下,匹配对数可增长23.07倍,可获取更密集可靠的同名像点.