以Subspace浮水印技術為基礎之數位音訊浮水印

Abstract

隨著資訊蓬勃發站及多媒體時代的來臨，數位內容傳播也越來越方便，使得智慧財產權和網路傳輸安全的問題與日俱增，而使用者對於網路多媒體的需求與日俱增，不可忽視。因此數位音訊浮水印的發展已刻不容緩。 本論文提出一個以subspace浮水印技術為基礎的數位音訊浮水印系統，藉由在DCT轉換後的頻率域上事先統計並分析可能遭受Stirmark Benchmark、LAME MP3壓縮等各各樣攻擊的原始訊號，進而分析出訊號對於這些攻擊較不易被破壞的特徵範圍，並將浮水印訊號嵌入在這些特徵範圍內。 實驗結果顯示此浮水印系統對於這些攻擊的強韌性足夠，並且能夠符合高偵測機率與低誤警機率的目標。

In recent years, due to the widespread use of Internet and great progress of the computer technologies, the proliferation of digital multimedia or documents is more and more important and can not be disregarded. To protect the intellectual property of the digital data, there has been significant interest in watermarking in recent days. We proposed a watermarking strategy from in which the watermark of a host audio data is selected from the robust features of the estimated forged audios of the host.

In this thesis, the robust features will be selected from the coefficients of the low frequency part of DCT coefficients after the DCT transformation. The forged audios are obtained from some software, such as LAME compression encoder and Stirmark Benchmark for Audio (SMBA). They perform potential pirate attacks on the host audio. We apply SVD technique on the features of the forged audios and obtain two orthogonal spaces. One of them characterizes most of the variations in the modifications of the host. We embed our watermark in the other space that most potential pirate attacks don’t touch. Thus, the embedded watermark is robust. According to the result of some experiments, it shows that our watermarking method can reach the goal of both of high detection probability and low false-alarm probability.