摘要
介绍了SIFT特征的的生成,包括尺度空间的生成、空间极值点检测、精确确定极值点位置、关键点方向分配和特征点描述子生成等5个步骤;其次针对航空像片的成像特点,给出了一种具体的特征匹配方法:在一个几何模型的约束下,应用可靠的匹配点来扩充其他的匹配点,然后多次应用RANSCN方法去除匹配精度较低的点.最后,给出了3组具体的航空像片进行实验.实验结果表明:不论是森林、居民地,还是河流,用这种方法都能找到很多特征点.但是,对重叠区域的匹配点进行分析得出:匹配点的多少不仅与重叠区域的大小有关,而且与影像所处的位置有关,对于特征比较明显的河流、居民地等地区,能找到很多匹配点,匹配成功率高;但对于特征不明显的森林地区,能找到的匹配点少,匹配成功率低,有些匹配点有明显的错误,这时可以通过手动或平差的方法去除匹配错误点.
In this paper, SIFT features which have obtained a success in the panoramic image matching are used in the air - photos matching problem. Firstly, generation of features of SIFT（ Scale Invariant Feature Transform） is introduced, which contains five steps： generation of scale - space, cable - space extreme detection, accurate key point location, key point orientation assignment and key point descriptor representation. Secondly, in view of the imaging characteristics of air - photos, a specific feature matching method is given： reliable match points are expanded to get more matching points, and the points with lower accuracy are removed by using RANSCN repeatedly. Finally, by making an experiment with three groups of specific air - photos, whether it is forest, residential area, or river, this method can find a lot of feature points. However, analyzing the matching points of the overlapping region shows that the number of matching points is related with not only the size of the overlap region, but also the locations of the image, and more match points will be found in the obvious characteristics image, such as river, residential area and so on with a high success rate, whereas the indistinct forest areas only can find less matching points with low success rate and clear mistake, and at this time, the wrong point can be removed manually.
出处
《重庆工学院学报（自然科学版）》
2009年第9期110-114,共5页
Journal of Chongqing Institute of Technology
关键词
自动匹配
SIFT特征
航空像片
特征点
automatic matching
scale invariant feature transform feature
air-photos
feature point
