隨著遙測科技的發展，衛載影像的技術日趨進步，也廣泛的應用在舉凡土地利用、變遷偵測、水資源等不同的領域。另一方面，由於台灣位於板塊交界處，地震頻傳，變遷偵測便成為地貌變化、環境保護的重要依據，我們可以透過兩張不同時間相同區域的衛星影像來做偵測及校正，並將校正後的成果用在變遷偵測上，藉此提升校正的應用範圍。事實上，每一張衛星影像在大氣、地貌、溫度等均有差異，如何能提升多張影像之間的辨識度便成為一個重要的議題。 本論文建立在色彩轉換的基礎上，不使用複雜的大氣參數，對不同時間點的影像進行相對校正。文中先對影像前處理，包含融合影像、提高解析度，將每張影像互相匹配，接著我們使用色彩轉換以及白化-反白化的方法，藉由影像本身的特徵，將選取的資料做校正；最後，得到校正的結果之後，為了讓使用者能夠得到應用在變遷偵測上的差異，用影像差分進行變遷偵測，最後把校正與偵測分類，進而比較與分析，得到最佳的判斷方法。 我們將影像資料分成山區以及平地混和，使用的是921集集地震中橫公路路段以及雲林縣草嶺地區的衛星影像，並由分析數據得知，在不同的影像資料下，我們的影像校正演算法都能降低大氣擾動對影像的影響，在結果影像都有良好的表現。 Because of the development of remote sensing technology, space-borne images have improved their resolution in the past decades. Remote sensing has many applications, such as land management, change detection, water resource and so on. Moreover, Taiwan is in the divergent boundaries. There are many earthquakes every year and some cause serious landslide. Change detection by remote sensing is an efficient approach to detect terrain features’ change and for environmental protection. We compare to images of the same region but collected at different times to detect the changes. However, images are diverse not only on ground condition, but also atmospheric conditions. How to increase the recognition rate between images is an important issue. In this thesis, we reduce the atmospheric conditions based on color transform algorithm without using complex atmospheric parameters. Because of its simplicity, relative correction is commonly used recently. We first apply image fusion to enhance the resolution and match our images. Then, we adopt color transform and Whitening/Dewhitening method for correction based on the statistic of images. For applying the result in change detection, we use univariate image differencing to detect the difference between the images. Finally, quantitative analysis is conducted for performance comparison. The image scenes used for experiments are in Central Cross-Island Highway of Taiwan during 921 chi-chi earthquake and Cao-ling area in Yunlin county. From the result, our method can reduce the atmospheric disturbances on the satellite images. It provides a procedural for correction algorithms and yields good quality and performance.
