資料隱藏(Data Hiding)的技術可以隱藏資料於載體內，來達到通訊保護之目的。因此它提供一個安全方式，使得機密資料可以在公開或開放的管道中傳遞。向量量化編碼(VQ)的資料隱藏技術是影像資料隱藏的其中一種方式，它透過向量量化編碼技術將資料嵌入偽裝的影像中，因此可以避免秘密通訊被第三者攔截。而這類型技術的特點是進行通訊的雙方都需要透過同一本編碼簿，才能正確的對訊息編碼與解碼。向量量化編碼的資料隱藏技術可以是「向量量化編碼可回復」(VQ-Oriented Reversible)，而這種技術可以將偽裝影像還原到原始影像的向量量化的形式。從過去相關文獻來看，向量量化編碼可回復資訊隱藏技術進行的研究可以分成兩部分：以編碼串為輸出的技術以及以影像為輸出的方式。 本論文在這兩方面都提出新的技術來改善向量量化編碼可回復資訊隱藏技術的效能。論文第一部分是針對編碼串為輸出的技術進行研究探討，而本論文所提出的ASCM方法的貢獻在於它比現有的技術具備更高的壓縮率以及更高的藏量。論文第二部分是針對影像為輸出的技術進行研究探討，而本論文在這方面共提出兩種方法：Two-level coding方法與Greedy-USBIRDS方法。其中Two-level coding 有較好的資訊藏量而Greedy-USBIRDS有較好的偽裝影像品質與不錯的資訊藏量。第二部分的貢獻在於，本論文所提出的方法比現有方法有更高的偽裝影像品質，同時也具備高資訊藏量。因此從實驗數據來分析比較本論文的研究成果與其它類似的方法的成果，我們發現本論文的研究成果在編碼串輸出的技術或是影像輸出的技術方面都具有重大的突破，編碼串壓縮率平均可以達到0.439，在影像品質平均可以高達37.27dB，且平均藏量可高達42萬位元。 Image data hiding is one of the most important techniques for data and communication protection. It embeds data into an image carrier and produce a corresponding stego-image that can then be decoded to obtain the hidden data. Therefore, it provides a secure way to distribute data through a public and open channel. VQ-based data hiding is an image data hiding technique that focuses on the problem of using a shared codebook between the message sender and the receiver to embed data into a cover image and to extract the hidden data from the stego-image. VQ-based data hiding methods can be VQ-oriented reversible, referring to the ability to recover the quantized cover image from the stego-image. VQ-based data hiding methods can be classified into two groups according to their outputs: codestream as the output and image as the output. This dissertation has made contributions to both groups of the VQ-based data hiding methods. First, this dissertation presents a new method called ASCM that uses codestream as the output. It achieves better compression rate and higher embedding rate than prior similar approaches. Second, this dissertation presents two steganographic methods, the two-level embedding method and the greedy-USBIRDS method, which use images as their outputs. The former method achieves larger embedding capacity than the later method, and the later method achieves better stego-image quality with good embedding capacity. When compared with prior similar methods, both the proposed methods achieve better visual quality and larger embedding capacity on average. The experimental results indicate that, the proposed methods in this dissertation make significant contributions to the area of VQ-based steganographic methods in terms of applicable measures such as the stego-image quality (PSNR: ~37.2dB with embedding capacity: 423,283 bits), the ability to handle complex cover images, and the compression rate (Bit rate: ~0.439).
