| Summary: | 碩士 === 國立高雄應用科技大學 === 土木工程與防災科技研究所 === 100 === The core work of the Geospatial Information is to derive the earth surface 3D
information from the stereogram of EO(Earth Observation) satellite and to rebuild the
DEM(Digital Elevation Model). The critical technology of the image registration still
uses traditional area basic matching to process pixel identification, point detection, and
position measurement. In order to achieve automatically points detection and reach the
high precision of sub-pixel, a pack of sufficient feature points of the image are the
requirement to process image registration and ground truth position, etc.
With the fast development of the remote sensing, there are more and more images
can be gathered by multi-period, multi-source, and multi-resolution. However, these
images are faced the difficulty of image registration due to displacement, rotation, scale,
topology, and panorama of the images. In order to minimize the affine distortion of a
pair of images which are affected by displacement, rotation, scale, topology, and
panorama, these tow images have to find a pack of even distributed, sufficient, high
precise points. Therefore, a better image registration method can not only minimize the
differential of the scale and rotation but also improve the efficiency and resolution of
the image registration.
This thesis addresses the optimal strategy of automatically detection for control
points is using the feature matching and are matching of the Voronoi-Delaynay method
to automatically predict the position points by point. To automatically derive feature
points in different period and different source images, the conjugate principal points and
quote the feature matching algorithm is provided to analyze the differential of the
feature points. The resolution of the sub-pixel can be reached and the results can be the
reference for next images.
|
|---|
