| Summary: | 碩士 === 國立臺灣大學 === 工程科學及海洋工程學研究所 === 105 === Remote sensing image registration is still a challenging task because of the variety in image types and the lack of a consistent transformation. To improve image registration for remote sensing, a robust and accurate method is developed in this thesis. To begin with, a modified scale-invariant feature transform (SIFT) method is proposed for feature point detection and pair matching. Based on the properties of matched pairs, the standard grouping boundary (SGB) and confidence elliptical boundary (CEB) are computed for further examination. The SGB is used to categorize matched pairs according to the slope values. The CEB is a geometric contour in spatial scale to remove outliers if the key-point locations are outside the contour. At last, random sample consensus approach is implemented to find the most appropriate transformation function. The developed algorithm is tested on multi-temporal remote sensing images to show the variation in different timescales. The experimental results show the improvement of matching performance, the accuracy, and robustness of proposed method.
|
|---|
