摘要
当前高分辨率多光谱遥感图像自动配准的主要困难,在于图像特征提取的自动化程度不足,计算速度较慢。针对不同传感器和不同时空分辨率的高分辨率多光谱遥感图像,改进点特征的提取方法,获得较高精度和较快速度。首先构建三维高斯差分尺度空间,由低层获得粗匹配点,在空间上向高层索引特征点。然后,通过逐层搜索获得精匹配点。在各层中,通过特征点方向描述子的空间增强,提高特征点的质量和数量。最后,综合多光谱遥感图像的可见光和近红外等波段的同名点集,获得亚像素级的匹配点。试验了环境星、TM、GEOEYE、无人机遥感图像等高分辨率遥感图像,对改进算法结果进行了较全面的对比分析。
A modified approach of characteristic point extraction was presented, in order to improve the low automation and slow calculation of registration for high resolution and multi-spectral remote sensing images. This method can be used for different sensors with different temporal and spatial resolution, and obtain the better precision and faster computation. Firstly, based on the original SIFT algorithm, a new dimension was applied for progressive down sampled image spatial resolutions with two dimensional difference of Gaussian scale space. Thus, the three dimensional difference of Gaussian scale space was constructed for the gray image of individual bands. The coarse corresponding points, which were obtained from the lower layer, can be used for the spatial indices of the higher layers of the image. Then, the refined corresponding points were collected by the layer by layer searching. In every layer, since the spatial enhancement of the feature descriptors, the quality and quantity of the characteristic points were effectively increased. Moreover, the sub-pixel corresponding points were generated and integrated by the multi-spectral bands of the remote sensing image, such as the common corresponding infrared and visible bands. The remote sensing images of HJ, TM, GEOEYE and UAV were tested and compared with thisalgorithm. The results indicate that and multi-spectral remote sensing extraction method. this improved approach of automatic registration for high resolution images is more accurate and faster than original characteristic point
出处
《红外与激光工程》
EI
CSCD
北大核心
2012年第12期3285-3290,共6页
Infrared and Laser Engineering
基金
科技部国际科技合作计划(Y0I0010062)
中国科学院国际合作项目(GJHZ1003)
中国科学院对外重点合作项目(Y0Y00630KX)
关键词
自动配准
同名点
高斯金字塔
差分尺度空间
多光谱遥感图像
automatic registration
corresponding points
Gaussian Pyramid
difference scale space
multi-spectral remote sensing images
