差值擴張可逆式浮水印技術之性能提升

Improvement of DE-Based Reversible Watermarking

方杏鑫(Sing-Sin Fang)

指導教授 ： 吳昭明

國立虎尾科技大學/工程學院/航空與電子科技研究所/碩士(2012年)

摘要

隨著網際網路的快速發展，網路上有許多非法的使用者會利用網路安全性的漏洞來竊取或竄改傳輸中的重要訊息，使得訊息在傳遞過程中的安全性議題逐漸被重視。數位浮水印技術為一種機密資訊保護、秘密通訊及影像驗證等功能的技術，經由在原始影像上嵌入機密資訊成為浮水印影像，嵌入的機密資訊可以是影像原創者的驗證碼，當影像遭到竄改或是不法的使用時，利用驗證碼就可以證明所有權或是對不法使用者提出法律上的舉證。一般數位浮水印雖是人類視覺不容易發現的，但對於原始影像而言仍具有不可避免的失真存在，且此失真往往無法完全的移除，對於珍貴且重要的影像而言，此一特性成為使用數位浮水印之嚴重絆腳石。可逆式浮水印的提出有效的克服此一困難。為了在接收端可以將浮水印嵌入所引起的失真完全移除，使原始影像完整重現，嵌入端必須嵌入附屬資訊，以確保影像可以無失真的還原。此外，如何提升浮水印的嵌入容量也是一項重要的課題。
本論文基於Thodi等人的PE2演算法提出三種改善機制，經過本論文改善後，可提供高於PE2演算法架構之最大可負載容量以及浮水印影像品質。首先溢位/欠位平移演算法將PE2演算法中不可擴張的預測差值轉換成可擴張，提升了最大負載容量，而一張影像在PE2架構中不可擴張配對的數量越多，套用本方法後所實現的性能將是越有效的。在溢位/欠位平移演算法提升最大負載容量的同時，本論文提出兩種套用溢位/欠位平移演算法後控制其浮水印影像品質的機制，稱為影像失真限制(1)與(2)，此機制將套用溢位/欠位平移演算法後會造成失真過大的像素配對排除，使浮水印影像維持一定的影像品質，此舉確保在低負載容量時，影像品質不會為了增加負載容量而犧牲。而在影像中的不可擴張配對中經由溢位/欠位平移演算法平移後，再加上奇數預測值嵌入法，在預測值為奇數時不強制取偶數，直接利用原始的預測值進行差值擴張，預防因為強制取偶數所產生的失真出現。
實驗結果證明本論文提出之改善方法確實可以在有效提升最大負載容量的同時保持一定的影像品質，甚至提升影像品質。在交叉實驗結果中，奇數預測值嵌入法、溢位/欠位平移演算法與影像失真限制全部套用在PE2架構，比較影像失真限制(1)與(2)的效能，實驗結果4.2.5顯示出影像失真限制(1)由於沒有(2)方法中的直方圖平移產生的影像失真，故直接的最大可能失真限制可帶出較好的失真限制性能，但兩種方法則同樣可以得到最大負載容量及影像品質同時的提升。

關鍵字

可逆式浮水印 ； 預測差值擴張 ； 溢位/欠位平移演算法

並列摘要

Digital data can be transmitted easily nowadays due to the rapid development of multimedia technology and the dramatic expansion of the Internet. Thus, the issue of copyrights protection and security of digital data are important. Reversible watermarking scheme is proposed to solve these issues. Reversible watermarking scheme not only extracts the secret data but also recovery the original image. The property of recovery the original image from watermark image is necessary for some applications, such as medical, military and legal do-mains.
In this thesis, three improving mechanisms for prediction error expansion embedding are proposed. The first improving mechanism is overflow/underflow (O/U) shift operation .The maximum true capacity of Thodi et al.’s PE2 reversible watermarking scheme based on prediction-error expansion (PE2) can be effectively increased by O/U shift operation. The O/U shift operation makes non-expandable prediction errors expandable. To preserve the invisibility of watermark embedding/extraction, the O/U shift operation has to be invertible. The increased maximum embedding capacity is equal to the number of non-expandable prediction errors. Note that no additional overhead is required for O/U shift operation. The second improving mechanism is distortion limitation. Two different algorithms for distortion limitation and discussed. They are denoted as DL(1) and DL(2), respectively. Although the maximum embedding capacity is increased by O/U shift operation, the quality of watermarked image is degraded due to the distortion introduced by expanding embedding on some pixels that is expandable after O/U shift operation is large. The distortion limitation algorithms can help us to select better overflow/underflow pixels (smaller distortion) for data embedding. Combination of O/U shift operation and distortion limitation, the watermarked image has better quality than PE2 scheme. In PE2, the prediction output is forced to be an even number such that all pixels are changeable and the embedded information bits can be extracted from the LSB of watermarked pixels. However, this even-number forced predictor maybe enlarges the distortion. The third improving mechanism releases the even-number limitation on the predictor. A modification on the expanding embedding algorithm is proposed for odd predicted value.
The experimental results show that the maximum true capacity of the O/U shift operation scheme is larger than Thodi et al.’s P2 algorithm. Under the same embedding capacity, the distortion limitation and the odd predictor embedding have the better image quality. DL(1) is superior to DL(2) because DL(2) needs additional histogram shifting but DL(1) is unnecessary. When we use O/U shift operation to increase the maximum embedding capacity both distortion limitations can enhance the image quality.

並列關鍵字

reversible watermarking scheme ； prediction errors expandable ； overflow/underflow shift operation

參考文獻

[1] J. Tian, "Reversible data embedding using a difference expansion, "IEEE Transactions on Circuits and Systems for Video Technology, vol. 13, pp. 890–896, Aug. 2003.

[2] D. M. Thodi and J. J. Rodriguez, "Expansion embedding techniques for reversible wa-ter-marking," IEEE Trans. Image Process., vol. 16, no. 3, pp. 721–730, Mar. 2007

[3] D. M. Thodi and J. J. Rodriquez, "Prediction-error-based reversible watermarking," IEEE Conf. Image Processing, pp. 1549–1552, Oct. 2004.

[4] A. M. Alattar, "Reversible watermark using difference expansion of triplets," in Proc. Int. Conf. Image Process., vol. 1, pp. 501–504, 2003.

[5] A. M. Alattar, "Reversible watermark using difference expansion of 611quads," in Proc. IEEE Int. Conf. Acoustics Speech Signal Process., vol. 3, pp. 377–380, 2004.