基於特徵統計之時軸同步機制
關鍵字 – H.264，視訊浮水印，區塊極性，序號調變，時軸同步。
本論文中我們利用區塊極性與序號調變來達到浮水印的嵌入，其中區塊極性是由4x4 DCT轉換且量化後的DC係數來決定，而區塊序號則是由4x4 DCT轉換且量化後，其所有AC係數量級的總和來決定。區塊序號的最低位元被當成調變參數，藉由修改量化之AC係數來改變調變參數，進而達到嵌入浮水印的目的。為了抵抗時軸同步攻擊，像是畫面刪除，畫面插入或是畫面順序調換，我們亦提出基於特徵統計的時軸同步機制，特徵統計量是經由統計視訊內容中的局部變異量或是特徵空間分佈的投影量而得到，同時將其當作額外資訊傳送到浮水印萃取端。在浮水印萃取端，藉由比對額外資訊與從接收視訊中取出的特徵統計量，來達到時軸同步。
模擬結果顯示提出的浮水印系統有良好的效能，而且萃取浮水印不需要原始視訊，同時浮水印演算法有較低的複雜度，可適合即時性的應用；而基於特徵統計的時軸同步機制，可促使浮水印系統更有效的抵抗時軸同步攻擊。
Index modulation for H.264 video watermarking and
temporal synchronization based on feature statistics
Keyword -- H.264; video watermark; block polarity; index modulation; temporal synchronization.
H.264 is a new advanced standard. The applications of video on Internet or wireless networks become very popular nowadays. However, these digital contents can be easily modified and copied by end users. Hence copyright protection, copy control and integrity verification has become important issues in recent years. Digital watermarking is a means of claiming ownership of a data source.
In the proposed system, block polarity and block index modulation are used to achieve watermark embedding. The block polarity is determined based on the nonzero quantized DC coefficient in each 4x4 integer DCT block. The block index is the pseudo-quantized block activity that is represented by the sum of magnitude of quantized AC coefficients. The watermark embedding is actually performed by the index modulation that will modify quantized AC coefficient values by a small amount to force the activity to be quantized into a specific region. For resisting temporal attacks, such as frame dropping, frame insertion, and frame transposition, we also propose a temporal synchronization method for video watermarking by matching feature statistics. The feature statistics are calculated by local variances or eigenvalues in video content and sent as side information. Temporal attacks can be detected by comparing side information and feature statistics that be calculated from the received video.
Simulation results show that the proposed method performs well and extract embedded watermark without the original video signal. Additionally, the algorithm is not very complex and appropriate for real-time applications. Based on the extracted feature statistics, the video watermarking system is more robustness against temporal attacks.