| 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 === 97 === Image registration is an important image preprocessing for image analysis and others application, the aim of image registration is to rectify the geometric difference / distortion of two or more images of the same scene due to different imaging conditions. In the past, several correction methods were proposed to correct those geometric difference / distortion. However, some disadvantages still exist: (1) non-fully automated, (2) feature matching method, (3) mapping function, (4) incorrect point elimination mechanism and (5) stop calculating mechanism.
This study cite the application of Thin Plate Splines(TPS) as mapping function of image registration process to give consideration to computational efficiency and overcoming the regional geometric distortion of image and adopt Voronoi-Delaunay diagram based of graph theory for control points selection.
This study proposes incorrect point elimination mechanism and stop calculating mechanism, through the experiments confirmed, that the mechanism proposed in this study can effectively eliminate incorrect point, and stop calculating at a minimum of control points are calculated to reach the required value of accuracy nested.
|
|---|
