Cloud covers present in optical satellite images generally limit the data usage and increase the difficulty of data analysis. Information reconstruction of could-contaminated images thus plays a fundamental step in data preprocessing. A method to accurately and consistently reconstruct information of cloud-contaminated pixels in multitemporal remote-sensing images is proposed in this study. By utilizing temporal correlation of multitemporal images, a patch-based information reconstruction that spatiotemporally segments a sequence of images into several patches with similar temporal variation is performed, and then information from cloud-free and high-similarity patches is cloned to their corresponding cloud-contaminated patches. In addition, a seam passing through homogenous regions is determined for a cloud-contaminated region to reduce radiometric inconsistency. A cloud-contaminated region is segmented into several patches and their corresponding cloud-free patches are determined by a quality assessment index, and the multi-patch information cloning is solved by an optimization process with a determined seam. These processes enable the proposed method to accurately and consistently reconstruct missing information in cloud areas. Qualitative analyses on sequences of images acquired by the Landsat-7 ETM+ sensor and a quantitative analysis on a simulated data are conducted to evaluate the proposed method. The experimental results the ability of our method to generate visually smoothing reconstruction results.
雲層遮蔽是光學式衛星影像上不可避免的問題,此問題限制了衛星影像的使用範疇以及增加後續資料分析的困難度,為利於後續應用的拓展,如何重建遮蔽區域下的地表資訊便成為一研究議題。本研究為了拓展衛星影像之後續應用,致力於移除衛星影像上之雲層遮蔽並重建遮蔽區域下之地表資訊,藉此提升衛星影像的可用性。針對此問題,本研究提出一新的解決方法,利用衛星影像的週期性,也就是多時期影像資訊,來做為資訊重建的參考來源,再以本研究所提出之方法準確且無痕地進行地表資訊重建。基於多時期影像上的高時空相關性,本研究提出了片狀式鑲嵌法演算法,其結合了影像之時空特徵分割以及最佳鑲嵌線之選取來對遮蔽區進行資訊重建。在挑選重建資訊的來源上,本研究提出的影像之時空特徵分割依影像中擁有相似時空變化特徵的位置進行分群,各群別再依品質評估指標自多時期資訊中尋找出最佳的回填資訊來源,而後續的資訊重建則依據挑選出的最佳回填資訊來源來做為重建資訊之參考依據。此外,為了無痕且準確地進行資料重建,本研究採用解算帕森方程式(Poisson equation)來對資訊重建進行最佳化解算,與其他相關研究不同的是本研究不直接使用雲層遮罩之邊界作為鑲嵌線,而另尋一最佳鑲嵌線使之通過影像均調區來降低輻射值的不連續性,並作為帕森方程式穩定的邊界條件,從而達到更佳的無縫鑲嵌成果。總括而言,雲遮蔽區域會先經由時空特徵分割步驟切割成數塊片狀區域,依據影像品質排序的成果各自挑選出最佳的填補資訊來源,再以計算選取所得的最佳鑲嵌線輔助解算帕森方程式,準確且無痕地進行資訊重建。本研究所使用的定量分析資料是Landsat-7 ETM+衛星影像,實驗成果顯示,不論是在準確度或者視覺上,本研究所提出的方法都有優異的成果。