利用浮水印技術做MPEG視訊的保真驗證與註解隱藏

Abstract

在本論文中提出了兩種驗證MPEG視訊完整性與保真度的方法，及一種在MPEG視訊中藏入註解資料的方法。在視訊驗證的第一個方法中，我們利用易碎浮水印的技術與在係數中建立偶同位(even-parity)的關係，驗證信號被嵌入離散餘弦轉換區塊(DCT block)的離散係數中。在視訊驗證的第二個方法中，我們利用半易碎浮水印的技術，驗證信號被嵌入在預先選定的係數位置中，及MPEG的移動(motion)向量中。藉由檢驗從視訊抽取出的浮水印信號，視訊的完整性與保真度得以被驗證。在所提出註解隱藏的方法中，我們利用強韌浮水印的技術，在一個8´8的離散餘弦轉換區塊中藏入一個位元的註解資料，同時在抽取註解資料的過程中，利用投票機制(voting scheme)決定最終抽取出來的註解資料。利用上述所提出的方法，藏入的資料對視訊品質所造成的影響可以降低。此外，所藏入的資料可以直接地被抽取出來，而不需要利用到原始的視訊。實驗結果證明了所提出的方法是可行的。

Two methods for authenticating the integrity and fidelity of a MPEG video, and a method for hiding annotations in MPEG videos are proposed in this study. In the first method for video authentication, a fragile watermarking technique is employed, in which authentication signals are embedded in the DCT blocks by setting up even-parity relations among the DCT coefficients. And in the second method for video authentication, a semi-fragile watermarking technique is employed, in which authentication signals are embedded in selected DCT coefficients and MPEG motion vectors. The integrity and fidelity of a video are authenticated by examining the watermark signals extracted from the video. In the proposed method for annotation hiding, a robust watermarking technique is employed, in which an 8´8 DCT block is used to hide exactly one bit of the annotation data, and a voting scheme is utilized in a data extraction process to decide the final extracted annotation data. The proposed methods can be employed to minimize the perceptual degradation of video quality caused by the embedded data. Furthermore, the embedded data may be extracted directly without using the original video sequences. Good experimental results prove the feasibility of the proposed methods.