隨著網路的迅速發展，數位相機及影像處理技術的成熟，不僅影像取得容易，也可以應 用影像處理技術作修改，影像造假(包含縮放旋轉、合成)也變得更容易，蓄意不良的造 假影像因而層出不窮，對社會大眾有許多不良的影響。面對這樣的問題，影像造假的偵 測技術更顯重要與迫切 本計畫主要提出一個有效的偵測影像重新取樣(resampling)的方法。對每個重新取樣倍 率，我們提出一個計算其零化濾鏡的演算法，此零化濾鏡與做了該倍率的重新取樣之影 像做旋積運算，會得到零的數值，故可以用來偵測影像是否做了該倍率的重新取樣。在 現有的方法裡[14, 15]，不是對倍率範圍有所限制就是只能偵測是否經過取樣而無法求出 倍率，而且計算時間較長。我們的方法初步用了45種零化濾鏡做偵測，倍率限制範圍比 這些方法放寬許多，最大為12倍。倍率限制可以利用增加不同的零化濾鏡來放寬。然而， 此方法用幾種零化濾鏡就只能偵測幾種取樣倍率，所以在第二階段我們試著利用有限個 零化濾鏡近似偵測更多不同的取樣倍率。初步的實驗已呈現不錯的結果。 With the widespread use of Internet and availability of powerful image processing and editing software, digital images are easy to acquire through internet and to manipulate and edit. Due to the fact that resampling is frequently involved in such manipulation, we would like to propose an effective method in this proposal to detect resampling in images. The proposed method includes three parts. The first part is an algorithm that automatically derives the coefficients representing the correlation between a resampled signal by any fraction and the original signal. The second part is an algorithm that uses these coefficients to construct a zeroing mask for the resampling by the corresponding fraction. The last is an algorithm that uses the zeroing masks to detect resampling on images. Compared with the existing methods, our proposed method for resampling detection can not only effectively detect resampling by any fraction but also figure out the resampling fraction. Furthermore, we use a set of 45 zeroing masks to detect a much larger set of resampling factors. Preliminary experimental results show that the proposed method is indeed effective.
