摘要
本文提出了一种多重约束下由粗到精的多源图像自适应子像素级配准算法。该算法采用影像特征点作为匹配基元,利用具有不同精度等级的组合判据法、整体松弛法、最小二乘法实现由粗到精的匹配,同时在匹配过程中加入了多重约束,如定位点控制约束、交叉匹配约束、连续控制约束,以保证获取的配准控制点的可靠性和剔除粗差点。此外,该算法利用配准控制点自适应地构建整个图像的三角网,最后依据改进的三角形填充算法对目标图像进行逐像点纠正。对同源和非同源的遥感图像的实验证明,SPOT4全色图像(10m/pixel)和SPOT5多光谱图像(10m/pixel)的配准精度分别达到6~7m和5~6m。
A multi-sensor remote sensing image sub-pixel registration algorithm was proposed in this paper. In this registration scheme, point feature, three image matching algorithms (ICM, PRM, LSM) and three constraints (APC, CDM, TCC) were applied to guarantee that the accuracy of point matching reached sub-pixel and that the distribution of feature points was self-adapted to the area terrain. Then Triangulated Irregular Network (TIN) was constructed with feature points. Finally the target image was rectified based on TIN with the improved triangle filling algorithm. Experiments results on multi-sensor remote sensing images demonstrate that the registration accuracy of the SPOT4 panchromatic image (10m/pixel) is 6-7 m and that the SPOT5 multi-spectral image (10m/pixel) is 5-6 m.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2007年第6期57-66,共10页
Opto-Electronic Engineering
基金
国家863计划项目(2002AA783050)
河南省杰出青年科学基金项目资助
关键词
图像配准
多重约束
三角网
图像处理
Image registration
Multi-constraint
Triangulated irregular network
Image processing
作者简介
邢帅(1979-),男(汉族),河南信阳人,博士生,主要研究方向为遥感影像处理与分析、数字摄影摄影测量、多源遥感数据融合. E-mail:xing972403@163.com.