餘弦轉換統計模型在音訊浮水印之研究

Abstract

理想的浮水印系統設計應兼顧其隱密性與強軔度，前者要求嵌入的認証訊號不能被盜版者察覺其存在，而後者則強調非法的惡意攻擊不致於破壞浮水印的完整性。一具體可行的方案是應用離散餘弦轉換配合展頻調變原理來建構浮水印的嵌入函數，其關鍵在於設定一能涵蓋音訊轉換係數特性的機率模型，同時依據統計預估原理設計一最佳的浮水印檢測及解碼機制。目前相關研究採用的相關式檢測器，原本依據高斯機率模型的假設而推導，並不足以正確反應不同演奏內容音樂的頻域係數分佈變化。有鑑於此，我們將先量測音樂頻域係數的統計特性，配合廣義高斯機率函數的數學推導，重新規劃其快速實現的最適化浮水印檢測及解碼機制。

In this thesis, we investigate an additive spread spectrum watermarking system in the discrete cosine transform (DCT) domain. The embedded watermark must be imperceptible by the user and should be robust to various signal manipulations. The detection problem can be formulated as a binary hypothesis test with the aim at maximizing the probability of detection conditioned to a given probability of false alarm. Furthermore, the decoding performance can be measured in terms of the probability of error. Most current research concentrate on correlation detectors, despite evidence showing that the underlying Gaussian model assumption does not match the intrinsic natures of DCT coefficients. Recognizing this, we will propose a statistical approach to model the DCT coefficients of the audio and then use it as a basis for the application of statistical decision theory to the design of efficient detection and decoding structures.