基於差值擴張之簡易無失真資訊隱藏技術

SIMPLE DE-BASED LOSSLESS INFORMATION HIDING TECHNIQUE

趙品竣(Pin-Jun Zhao)

指導教授 ： 吳昭明

國立虎尾科技大學/電機資訊學院/光電與材料科技研究所/碩士(2010年)

摘要

可逆式資料隱藏技術被提出不僅將隱藏資料嵌入與取出而且可以由偽裝影像還原至原始掩護影像。
Jin等學者提出區塊式可逆式浮水印方法。(根據Tian的架構)使用差值擴張的方法將隱藏資料嵌入於每個影像區塊。利用影像區塊中心像素值與周圍八點像素值的平均值形成差值擴張(DE)必要的像素對。雖然Jin所提出採用區塊式互相重疊的方法，隱藏資料嵌入的容量仍然很小。
在這篇論文中，我們提出兩種方法來改善Jin所提出嵌入容量的方法。Jin所提出這個方法是將一張掩護影像分割成3x3影像區塊大小。但是我們所提出兩種方法，是將掩護影像分割成3x3互相不重疊的影像區塊大小。
我們提出第一種方法，利用每個影像區塊角落四個頂點來嵌入隱藏資料位元。使用Tian等人所提出差值擴張(DE)嵌入方法。在這影像區塊利用其它五點像素值來預測角落四個頂點像素值。當接收端取出隱藏資料時這五點像素值不會被改變。利用原始像素值和預測值之間的像素對形成差值擴張(DE)嵌入方法。我們定義一個新的參數為正規化標準差來搜尋門檻值。在接收端利用正規化標準差來區別可擴張和不可擴張的像素對。理想的情況下，每個3x3的影像區塊可以嵌入四個隱藏資料位元。所以，這個新的方法嵌入隱藏資料容量比Jin所提出方法更高。
為了進一步提升嵌入容量，我們提出第二種方法不僅是影像區塊角落四個頂點而且更使用中心像素值為了讓隱藏資料的嵌入。理論上，每個3x3影像區塊可以嵌入五個隱藏資料位元。在嵌入過程中其它四點像素值不會被嵌入隱藏資料所以不會被改變。我們所提出這兩種方法能嵌入隱藏資料容量比Jin所提出方法更高。

關鍵字

區塊式差值擴張可逆式浮水印

並列摘要

Reversible information hiding technique is proposed not only for data embedding and extraction but also for recovery of the original cover image from the stego-image.
Jin et al. proposed a block-wise reversible watermarking method. The difference expansion (DE) data embedding method (proposed by Tian) is used for each image block. The necessary pixel pair for DE is formed by the center pixel and average intensity of the surrounding 8 pixels in the same image block. Although overlapped block partition is adopted in Jin’s scheme, the capacity of data embedding is still small.
In this thesis, two schemes for improving the embedding capacity of Jin’s method are proposed. The same as Jin’s scheme, the cover image is partitioned into 3x3 image blocks. But, in both proposed methods, the image blocks are non-overlapped.
In the first proposed scheme, the 4 pixels around the four corners of each image block are used to embed secret data bits. Tian’s DE embedding method is used. The other 5 pixels in the same block are used to predict the pixel intensities around the four corners. These 5 pixels remain unchanged for data extraction. The original pixel intensity and the corresponding prediction intensity form a pixel pair for DE embedding. We define a new parameter named as normalized standard deviation for the search of threshold. This threshold is used for distinction between embeddable and non-embeddable pixel pair in data extraction process. Ideally, 4 secret data bits can be embedded into a 3x3 image block. So, the embedding capacity of this new scheme is higher than Jin’s method.
To increase the embedding capacity further, not only the 4 pixels around the four corners but also the center pixel are used in the second proposed scheme for data embedding. Theoretically, 5 secret data bits can be embedded into 3x3 image block. The other 4 pixels in the same block remain unchanged in the embedding process. The embedding capacity of these two schemes is higher than Jin’s method.

並列關鍵字

prediction-error expansion ； normalized standard ； threshold

參考文獻

[4] C. De Vleeschouwer, J. F. Delaigle, and B. Macq, “Circular interpretation on histogram for reversible watermarking,“ in IEEE Int. Multimedia Signal Process. Workshop, France, Oct. 2001, pp. 345-350.

[6] J. Fridrich, M. Goljan, and R. Du, “Lossless data embedding-New paradigm in digital watermarking,” EURASIP J. Appl. Signal Process., vol. 2002, no. 2, pp. 185-196, Feb. 2002.

[9] Celik, M.U.; Sharma, G.; Tekalp, A.M.; “Lossless watermarking for image authentication: a new framework and an implementation Image Processing,” IEEE Transactions on Volume 15, Issue4, April 2006 Page(s):1042-1049.

[11] Alattar, A.M.; “Reversible watermark using difference expansion of triplets Image Processing,” 2003. ICIP 2003. Proceedings. 2003 International Conference on Volume 1, 14-17 Sept. 2003 Page(s):I-501-4-vol.1.

[13] Thodi, D.M.; Rodriguez, J.J.; “Expansion Embedding Techniques for Reversible Watermarking,” Image Processing, IEEE Transactions on Volume 16, Issue 3, March 2007 Page(s):721-730.