具回塑性及無失真資訊隱藏之研究

Abstract

隨著電腦科技與網路技術的發展，資訊隱藏技術的發展更加多樣化，利用不同載體（Cover media）（如：圖片、音訊及文字檔案等）的資訊隱藏技術，不斷的推陳出新。由偽裝媒體（Stego media）內容是否失真（即偽裝媒體與載體內容是否一致），資訊隱藏技術可分成失真及無失真（Lossless）兩大類。失真類的資訊隱藏技術，依據是否具備回復性（Reversibility），即原始載體是否可由偽裝媒體完美的回復，又可分為可回復性（Reversible）及不可回復性（Irreversible）兩類。近來，由於醫學、軍事及法庭科學等應用領域，載體內容不容有所失真，可回復性的資訊隱藏技術變得越來越重要。 本論文共提出兩種可回復性及一種無失真的資訊隱藏技術。首先提出以數位影像為載體的可回復性資訊隱藏技術，在此技術中，我們應用邊緣預測方法（Edge-oriented prediction），進行倍增誤差（Prediction error expansion）的資訊隱藏。實驗結果顯示，透過邊緣預測方法，可得到較小的估計誤差，因此在相同的資訊隱藏量下，可使偽裝影像達到較小的失真。 其次，我們提出以立體影像（Stereo image）為載體的可回復性資訊隱藏技術。由於立體影像的構成，來自成對不同位置之攝影機所拍攝的數位影像，成對影像中存在許多相似的影像區塊，我們將資訊隱藏於這些成對的相似影像區塊中。首先將兩張數位影像進行數位餘弦轉換（Discrete Cosine Transform），依據影像區塊之低頻係數，找出兩張影像中差異最小的相似影像區塊，之後計算每對相似影像區塊之中頻係數的差異，再利用雙向的直方圖位移(Histogram shifting)資訊隱藏技術，將訊息藏於相似影像區塊之差異值為0中頻係數。經實驗結果，藏入訊息的偽裝立體影像可維持極佳的影像品質，並且有極高的資訊隱藏量。 最後，我們提出以簡報檔案（PowerPoint file）為載體，利用簡報檔案中的特效，進行無失真之資訊隱藏。在學術或商業簡報中，常運用簡報檔案中的特效，協助主題的聚焦或時間的控制等。因此，利用簡報檔案中動畫或時間等特效來進行資訊隱藏，不但不影響簡報檔案的內容正確性，更可豐富簡報的呈現與吸引力，亦可抵抗檔案轉換攻擊。經實驗結果驗證，我們提出的方法可以抵擋若干視覺上或統計上的攻擊。

Steganography is a research topic to protect the privacy communication. The techniques can be divided to two categories: lossy steganography and lossless steganography. Lossy steganography hides message in the content of cover media and distorts the real content. On the other hand, lossless steganography hides message in the cover media and keep the real content intact. For some applications such as medical images, military images, or forensic images (provided by law), reversible property, where the cover image can be perfectly recovered from the stego image, is an important issue. Therefore, lossy steganography can be divided into two categories: reversible steganography and irreversible steganography. Reversible steganography can perfectly recover the cover media in the extraction process, whereas irreversible steganography cannot. In this dissertation, we will propose two reversible and one lossless steganographic methods using different cover media. First, a difference expansion based reversible steganographic approach, which yields low distortion under the same embedding capacity, is proposed. An edge-oriented prediction method and a modified overflow/underflow prevention method are proposed to achieve the requirement. Experimental results have demonstrated that the proposed reversible steganographic method yields lower distortion than several well-known expansion-based reversible steganographic methods. Second, a novel DCT-based reversible steganographic method for stereo images is proposed. Stereo images captured from a pair of CCDs simultaneously are widely used to create the illusion of 3D depth. Every pair of stereo images has many similar block pairs. In this method, we embed secret data in those similar block pairs. Experimental results have shown that the proposed method is undetectable under Chi-square analysis and outperforms other methods in terms of embedding capacity and image quality. Finally, we propose a lossless steganographic method to embed message in a PowerPoint file. In the proposed method, we not only hide message naturally but also keep the content of the cover media intact. Furthermore, the proposed method can resist the format conversion attack. The experimental results demonstrate that the proposed method is undetectable under some visual and statistical attacks.