在本論文的研究中，我們提出一個以多時段SPOT衛星影像來達成雲層自動去除的方法。這個方法共分成三個步驟：多頻譜影像增強、決定雲與雲陰影區塊、及影像鑲嵌。在第一步驟中，我們將影像色彩空間由RGB轉換到YCbCr進行影像增強，再依序調整亮度分量（Y）及色彩資訊（CbCr）分量。在第二步驟中，我們利用線性頻譜分析法（linear spectral unmixing method，LSU）標示出可能屬於雲材質的區塊，再利用一系列修正方法去除被誤判的區塊；例如，建築物、平原等高亮度像點，並且在厚雲區塊周圍找薄雲和雲陰影區塊，以達成標示出最佳雲及雲陰影區塊的目標。在第三步驟中，我們使用多重解析度離散小波轉換方法（multiscale discrete wavelet transform method）來解決影像接合所造成的邊界不一致的問題。 Partial cloud cover is a severe problem in optical satellite remote sensing. This problem can be partially overcome by acquiring multiple images at different time over a given region and the reasonably cloud-free composite image can be obtained by mosaic of the cloud-free areas in the set of images. In this paper, a multidisciplinary operational algorithm is proposed to generate cloud-free mosaic images from multi-temporal images acquired by the SPOT satellite images. First, the original images were enhanced in the YCbCr space. Then we choose the base image that has the least thin cloud cover and divide the base image into several grid zones. In the cloud determination, we use the linear spectral unmixing method (LSU) to extract all cloud cover pixels, but the method cannot handle thin clouds and cloud-shadow, and often confuse bright land surfaces as clouds. Therefore, we utilize an opening operator to exclude smaller confuse bright zones and eliminate unnecessary marking like plain or large sized buildings by checking difference with another SPOT image. At last, we find thin cloud and cloud-shadow in eight-neighbor grid zones around thick clouds based on the geometric relation and sun elevation angle. The cloud and cloud shadow grid zones of the base image are replaced by the same zones of other images. Comparing the base image and replaced image, we create a transition zone. Finally, the multiscale wavelet pyramid method is used to blend images in the transition zone.
