本論文提出一個以多重區塊特徵提取的方式來偵測影像是否遭到複製移動竄改的攻擊，首先將影像分割出來的區塊進行不變動量值的特徵處理，再對不變動量值進行兩種不同特徵的提取，接著以這兩種特徵值對區塊進行分類，並且將每個類別集合成大類以增加區塊相似度的範圍，擷取在兩種特徵值分類中皆為同類別的區塊進行比對，如此降低區塊間的比對次數；另外，本方法利用不變動量值進行區塊比對，除能提升本方法對不同角度的偵測外，亦降低區塊的資料量，使得比對效率大幅提升，且能同時過濾極度不相似的區塊。在匹配階段，將偵測配對之區塊以記錄水平垂直距離的方式進行匹配，以找出被竄改的區域，並根據被竄改區域間的關係，分別出不同性質的竄改區域。實驗結果顯示，本論文在極佳的偵測結果下，同時提升整體的執行效率。 This paper identifies copy-move forgery regions in an image through invariant features extracted from each block. First, an image is divided into overlapped blocks and 7 invariant moment features of the circle area under each block are calculated. Two features, mean and variance, are then acquired from the 7 moment features in each block. Each block is only compared to those blocks under the intersection of the same mean and variance feature sets. The copy-move forgery regions can be found by matching the detected blocks with the identical distance. Moreover, the adopted moment features are efficient on detecting rotational blocks. Experimental results show that the proposed scheme detects rotational duplicated regions well.
