Please use this identifier to cite or link to this item:
Title: 影像特徵點萃取與匹配應用於福衛二號影像幾何糾正
Apply Feature Matching on FORMOSAT-2 Image Georectification
Authors: 張國楨
Chang, Kao-Chen
Chen, Chun-Kai
Keywords: 影像匹配
image matching
image feature
Issue Date: 2010
Abstract: 隨著數位影像獲取及電腦科技的快速發展,遙測科技應用逐漸將領域由國防軍事用途拓展到其他應用範圍,大幅的增加其應用性與時效性。影像幾何糾正為一不可或缺,且日益重要的基礎處理,關係著後續影像融合、變遷偵測或影像鑲嵌等應用。本文探討如何改進特徵匹配演算法來提升遙測影像幾何糾正的效能。尺度不變特徵轉換(Scale Invariant Feature Transform, SIFT)為一種針對高解析數位影像發展出來的影像特徵點萃取方法,其優點是所萃取的特徵不易受到影像旋轉、縮放和灰度值差異而有所變化、皆具有良好特徵點選取與匹配,其結果更可靠和消除影像處理中不確定性。但原始演算法並非針對遙測影像特性所發展,直接應用不易得到良好匹配結果。本研究針對遙測影像的特性,提出改善之方法並修改演算法、建立半自動化處理流程。以FORMOSAT-2 影像為範例,探討SIFT於遙測衛星影像適用性。研究成果可利用修改SIFT演算法於研究中針對多時期、不同區域、不同載具衛星影像進行影像匹配,得到足夠影像控制點,以其進行影像對位與影像幾何糾正,整體精度RMSE優於0.5 pixel。經半自動化處理流程,可將原本耗時、經驗人力導向幾何糾正工作大幅縮短所需時間。
With the digital image acquisition and the rapid development of computer technology, application of remote sensing technology by the defense and military fields gradually extended to other application purposes, a substantial increase in its applicability and timeliness. Image geometric correction for the integral and increasingly important foundation treatment. It related with the subsequent image fusion, change detection or video mosaic applications. In this paper, a robust feature extractor technique for FORMOSAT-2 image rectification is applied and discusses how to improve the feature matching algorithm to improve the performance of remote sensing imagery to correct. The Scale Invariant Features Transform (SIFT) method extracted features are computationally attractive and invariant to image rotation, scale change and illumination. The original algorithm was not developed for remote sensing image features, and the results of applied directly are difficult to get a good match. In this study, a modified SIFT method is proposed and create a semi-automated processes on the applicability of remote sensing satellite images. Used FORMOSAT-2 as an example of SIFT on the applicability of remote sensing satellite images. Modified SIFT algorithm tests on different times, different regions, and different set of satellite images with image matching, the results obtained sufficient image control points, and and the overall accuracy is better than any set of 0.5 pixel. The semi-automated processes can be simplified and Shortened the time required of original that time-consuming, empirical and labor-oriented geometric rectification work. The result got good performance, precise, and more reliable and removed uncertainty of image processing.
Other Identifiers: GN0891230034
Appears in Collections:學位論文

Files in This Item:
File SizeFormat 
n089123003401.pdf11.99 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.