论文部分内容阅读
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。
The main difficulty of automatic registration of high-resolution multi-spectral remote sensing images is the lack of automation of image feature extraction and the slow calculation speed. Aiming at high resolution multispectral remote sensing images with different sensors and different spatial and temporal resolutions, the method of extracting point features is improved to obtain higher precision and faster speed. Firstly, we construct the 3D Gaussian difference scale space, get the rough matching point from the lower level, and index the feature points to the upper level in space. Then, through the layer-by-layer search for fine matching point. In each layer, the quality and number of feature points are improved by increasing the space of the descriptor in the direction of the feature points. Finally, the sub-pixel-level matching points are obtained by synthesizing the same-name point sets of the visible and near-infrared bands of multi-spectral remote sensing images. The high resolution remote sensing images of environment star, TM, GEOEYE and UAV remote sensing images were tested. The results of the improved algorithm were compared and analyzed in a comprehensive way.