The Study of Image Matching and Registration for Remotely Sensed Images Based on Local Invariant Feature
| Autor: | Shiuan-Yu Lin, 林炫佑 |
|---|---|
| Rok vydání: | 2011 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 99 Using the image technology to acquire geographic data is the main task in Geospatial Information technology. To select some conjugate points in stereo image pair is the necessary procedure for three Dimensional model reconstruction, map drawing or image registration. In traditional image matching method, the area-based matching is a mainly approaching. Even if the area-based matching method can reach to sub-pixel level accuracy, but it will fail matching while different scale, rotation, illumination, homogeneity and texture repeatability between two images. In recent years, the feature-based matching method is developed quickly that can overcome some difficult in area-based matching and increase the efficiency of matching and accuracy. In this study, we introduced the current feature based algorithm with SIFT, SURF, ASIFT, MSER respectively to feature selection and matching in remotely sensed images and close-range photogrammetry. Expecting it can archive enough control points automatically in the image registration procedure. The research firstly to discuss the theory of these algorithms, then used a lot of various type images to experiment and analysis, including different time, scale, rotation, geometry images. Furthermore, the Thin Plate Splines (TPS) function using in image registration has introduced to eliminate the error points. The experiment results show the adaptability of each algorithm may be difference, but it can reach to sub-pixel level accuracy each method. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|