摘要
尋找有變化之區域一直是一個很重要的課題，而這個問題可以使用同一地點不同時間拍攝之影像序列來做出分析。在這篇論文裡面，我們回顧了一些像素導向及紋理導向之影像變化偵測方法，除此之外，我們開發了兩個變化偵測方法，並分別以衛星影像、無人機影像來做測試。我們開發的第一個方法為使用影像對的局部相關性(Pearson's correlation)以及回歸影像之局部直方圖的Kullback–Leibler距離來量測影像局部之相似性後做變化偵測，此方法的時間複雜度是線性的；第二個方法為使用Gabor濾波庫提取影像對的局部特徵，並對此用餘弦相似度做量測。最後對此取一個手選的閥值來得到二元變化遮罩。
並列摘要
Finding change parts is an important issue, and we can detect change areas by analyzing images of the same scene at different time. In this thesis, we review some pixel-based and texture-based change detection techniques. In addition, we developed two change detection methods, and we apply those methods on satellite and drone images. The first method, we measure the similarity between image pair by their local Pearson's correlation and histograms' Kullback–Leibler distance of the regressed image pairs. This method is implemented in linear order time complexity; In the second method, we extract features by Gabor filter bank, and measure the cosine similarity between features. A manual threshold is applied to get the binary change mask.